Abstract
The pharmaceutical industry suffered a significant decline of innovation in the last few decades, whose simple reason is complex biology. Artificial intelligence (AI) promises to make the entire drug discovery and development process more efficient. Here I consider the potential benefits of using AI to deepen our mechanistic understanding of disease by leveraging data and knowledge for modeling and simulation of genome-scale biological networks. I outline recent developments that are moving the field forward and I identify several overarching challenges for advancing the state of the art towards the successful integration of AI with modeling and simulation in drug discovery.
1 Introduction
The worldwide spending on pharmaceuticals (i.e., prescription drug sales) hit US 1.3 trillion in 2020 and is forecast to reach US 1.7 trillion by 2026 at a compound annual grow rate (CAGR) in the range 3–6% (IQVIA, 2021). Pharma companies invest on average 15% of their revenues in R&D, which is about five-fold more than the public expenditure, a trend that has run unchanged over the last 30 years (Prokop and Michelson, 2012). Nonetheless, the last 2 decades have seen a clear deterioration in productivity, with an approximately 50% reduction in peak sales per asset and a >80% decrease in return of investment (ROI) for the pharmaceutical industry during the years 2010–2019 (Hay et al., 2014; Steedman and Taylor, 2020). While the Covid-19 pandemic somewhat mitigated these figures, the cost of developing a new drug and bringing it to the market is still on the rise and it is now estimated at around US 2 billion (i.e., roughly doubled in the last 10 years) (May et al., 2022).
Such decline in innovation is easily explained by the complex, and often hidden, biology of diseases, which makes it difficult to identify and validate biological targets of physiopathological significance (Prokop and Michelson, 2012). Our lack of disease understanding is a major reason for the high failure rate of clinical trials, which stands at about 90% (but possibly underestimated by the successful development of “me-too” drugs) (Kola and Landis, 2004). Indeed, most drugs fail as a result of low efficacy (45–65% of cases) or high toxicity (15–35% of cases), and seldom because of strategic, commercial, or operative reasons (Harrison, 2016; Takebe et al., 2018; Dowden and Munro, 2019). Clinical trials have even occasionally had serious adverse effects on trial participants, including death or severe injuries like hepatotoxicity (e.g., fialuridine) or multi-organ failure (e.g., TGN 1412, BIA 10–2474) (Institute et al., 1995; Attarwala, 2010; Peck et al., 2022). Additional complications are sometimes due to unforeseen side effects of drugs already on the market that have therefore to be recalled for safety issues.
Given the recent advances in computational biology and bioinformatics, it seems a particularly good time to seek support from modeling and simulation, i.e., carry out virtual experiments using digital copies of human cells, tissues, and/or organ systems, especially when in vitro or animal models are notoriously unpredictive of human biology [for example, see (Booth et al., 2003)]. The ability to predict clinical efficacy and toxicity in silico can indeed be a time- and cost-effective solution for the pharmaceutical industry, thus empowering sponsors to take go/no-go decision earlier in the drug discovery and development process (Pritchard et al., 2003). Gaining information about a drug’s effects (including mechanism of action, biological targets, network interactions, and so on) is likewise important for small and medium-sized biotech companies as they are largely reliant on venture capital funding and are often unable to generate sufficient efficacy and toxicity data to avoid failures happening further down the line (Roberts, 2018).
2 Target-based vs. phenotypic drug discovery
Target-based drug discovery (TDD) is a top-down approach where the desired target is known (i.e., a protein thought to be relevant for a disease) and compounds are screened against their interaction with the target. Conversely, phenotypic drug discovery (PDD) is a bottom-up approach where the desired phenotype is known (i.e., healthy condition readouts) and compounds are screened against replication of the phenotype. Both approaches have the generation of chemical compounds as their endpoint, yet they are characterized by one profound difference. While TDD requires only the biological target for drug screening, PDD requires a full biological system, like cultured cells, animal models, or human subjects/patients. Such difference is very likely the reason for why in silico applications have found their main success in TDD. For example, computer-aided drug design (CADD) uses a variety of techniques, including quantitative structure-activity relationship (QSAR), molecular docking, ADMET (absorption, distribution, metabolism, excretion, toxicity), which can predict drug-target binding using feature descriptors, e.g., physicochemical properties (Jiménez-Luna et al., 2020; Jiménez-Luna et al., 2021; Paul et al., 2021; Galati et al., 2022; Göller et al., 2022; Jia and Gao, 2022; Mamada et al., 2022). Similarly, approaches based on artificial intelligence (AI) have been very successful in drug design, where generative models for in silico screening (for example, generative adversarial neural networks, GANNs) are capable of sampling the chemical space at an incredible pace compared with physics-based simulation such as molecular dynamics (Díaz et al., 2019; Prykhodko et al., 2019; Tong et al., 2021a; Blanchard et al., 2021; Gaudelet et al., 2021; Abbasi et al., 2022). On the other hand, our level of insight into the biological complexity of disease has been thus far deemed as insufficient for a fruitful in silico PDD approach.
Strategies for TDD are very popular due to their cost-effectiveness compared with PDD approaches, yet they carry the risk of having an incomplete preclinical validation and a lot of false positives. The problem is that the generation of a promising lead compound that successfully interact with a target does not immediately translate into the expected biological outcomes, because many targets are disease-associated (which is relatively easy to identify) but not disease-modifying [for example, see (Florez, 2017)]. In addition, many drugs, including most of the commercialized medicines, are pleiotropic (i.e., they interact with more than one target) and thereby interfere with several cellular processes including signal transduction, receptor-mediated signaling, and metabolic homeostasis. The ensuing changes in cell biochemistry stimulate feedback and feedforward loops acting either to amplify or suppress the activity of specific biological pathways within very complex networks. As a consequence, the final net biological outcome cannot be inferred from the individual effect of the drug on a given target. The relatively low relevance of target-specific effects is evidenced by the fact that for up to 20% of currently approved drugs the biological targets and mechanisms of action are actually unknown (Moffat et al., 2017).
The multiple interactions between drugs and biological targets might well underlie unwanted side-effects, but often they also contribute to therapeutical efficacy (Vincent et al., 2022). In particular, even mild modulation of several targets simultaneously can be much more efficient than strong modulation of a single target. This notion forms the basis of post-hoc polypharmacology (i.e., multi-target drug discovery and combination therapy), which illustrates the idea that it is not the target that should be treated, but rather the state of the biological network (van Hasselt and Iyengar, 2019). Arguably, there is an increasing demand on achieving a mechanistic understanding of biological targets that takes into account the fact that such targets are immersed in intricate networks and pathways often spanning multiple cell types (Arrell and Terzic, 2010; Benson, 2015). Given the high complexity and redundancy of biological networks, predicting how perturbing multiple targets affects cell biology, possibly at distinct points in space and time, requires the development of computational tools that are capable of capturing the holistic nature of living systems (Pujol et al., 2010; Rai et al., 2018). Specifically, in order to link TDD and PDD it is necessary that such computational tools implements target deconvolution, i.e., the identification of the biological targets that are responsible for a specific drug-induced phenotypic response (Hughes et al., 2021). Attempts in this direction have recently employed AI and static interaction networks [reviewed by (Cuperlovic-Culf, 2018; Zampieri et al., 2019)], e.g., compound-protein and phenotype-protein interactions (Iwata et al., 2020). However, as discussed below PDD and target deconvolution would greatly benefit from the integration of AI with dynamical modeling and simulation of genome-scale biological networks. Accordingly, cheminformatics, in silico target prediction, as well as pathway network analysis are thought to be critical tools along with the advancements in experimental drug target deconvolution (e.g., chemical proteomics) [reviewed by Hughes et al. (2021)]. This argument is especially important for multiple or non-protein biological targets, where computational methods provide prioritized recommendations for subsequent in vitro and in vivo experiments for target identification and validation (Leon and Markel, 2006).
3 Artificial intelligence in biological modeling and simulation
3.1 Overview
Systems biology aims at deciphering the complexity of biological networks based on the formulation and use of mathematical models. In particular, dynamical modeling and simulation investigates the behavior and the relationships between the elements of a particular biological system while it is functioning and responding to perturbations (Ideker et al., 2001). Typically, the dynamics of the biological network is modeled by a set of ordinary differential equations (ODEs) containing balance equations describing the time derivatives of chemical species as well as rate equations describing the velocities and mechanisms of action of chemical processes. The ODE-based approach represents one of the most detailed solutions for time-dependent analysis of biological networks (Tangherloni et al., 2017). However, it requires the knowledge of a large number of initial conditions and kinetic parameters, and until recently only relatively simple biological systems have been analyzed using such approach. Yet, being able to quantitatively interrogate how complex biological networks at the genome-scale (e.g., whole-cell level) respond to perturbations holds great potential for gaining critical insights into the mechanisms of diseases and the relevant therapeutic interventions (Kell and Goodacre, 2014).
Biological networks can be studied at various spatial scales spanning ∼10 orders (nm to m) of magnitude (e.g., atomic/molecular, molecular-complexes, sub-cellular, cellular, multi-cellular, tissue, organ, multi-organ, organism) (Sali, 2021). In order to cope with such complexity, multi-scale ODE-based models commonly neglect spatial heterogeneity, for example pools are considered instead of individual molecules, and/or populations instead of individual cells. Even with these simplifications, genome-scale models might still contain up to tens of thousands molecular species and reactions within different cells and cellular compartments, eventually across distinct body organs (Thiele et al., 2020). The interactions between system components are critically dependent on network architecture, or topology, which brings about emergent properties and behaviors so that phenotypic traits arise from the collective action of genes (and conversely, genetic diversity results in a variety of phenotypes). Collective phenomena include steady-state multiplicity (e.g., bistability), hysteresis, oscillations, whose details are outside the scope of this paper [for reviews, see (Goldbeter, 2018; Goldbeter and Yan, 2022)]. As for the relevance in drug discovery, the deterministic nature of ODE-based model simulation has the critical advantage of allowing the study of “emergence” in terms of quantitatively monitoring and analyzing the phenotypic response of the whole biological system to pharmacological, genetic, environmental, or pathophysiological perturbations.
3.2 Model construction
Genome-scale biological network models typically include gene-regulatory, signal transduction, and metabolic pathways. Metabolic models are among the most studied, as metabolism is an exquisite indicator of the homeostatic functional state of the cell and thus it is diagnostic of the patient’s disease phenotype (Angione, 2019). Reconstructions of biological network topology have recently advanced to the whole-body level, with genome-scale single-cell networks [e.g., (Brunk et al., 2018)] scaled to multi-cellular/multi-tissue systems (Thiele et al., 2020). Such a process is supported by large community efforts across research disciplines (Singla and White, 2021) that constantly generate petabytes of publicly accessible datasets from high-throughput biological experiments (Camacho et al., 2018). As highly complex multi-omics information (e.g., abundance of genes, proteins, and metabolites) are progressively more available, data-driven AI techniques become increasingly important for efficient model construction, validation and refinement.
Bottom-up biological network reconstruction is an iterative process that requires both well annotated genome sequences and literature-driven manual curation. For example, starting from raw whole genome/exome sequences, similarity-based (e.g., sequence homology or local alignment) and ab initio algorithms for gene-finding are used to predict the protein coding regions and the functional sites of genes (Wang et al., 2004; Ghorbani and Karimi, 2015; Chavali and Rhee, 2017). Recently, deep learning has been employed to annotate enzymes and capture functional difference of enzyme isoforms on the basis of raw sequence data (Li et al., 2018). Additional biochemical information is needed on a protein-specific basis (e.g., substrates and products for enzymes, or phosphorylation state for receptors), which is typically obtained using basic research (Reed et al., 2006). Other machine-learning algorithms predict the cellular localization of proteins from both the amino-acid sequence and high-throughput experimental approaches (Schneider and Fechner, 2004; Nakai et al., 2007). A combination of automatic and manual gap analysis (e.g., due to genome misannotations) is then required to obtain gap-filled biological networks (Karp et al., 2018; Pan and Reed, 2018). Finally, gene-expression profiling via transcriptomic data or reference atlases (Lähnemann et al., 2020), possibly with the help of machine learning (ML) (Abdolhosseini et al., 2019), is utilized to prune the biological network in order to generate the single-cell reconstruction (Raškevičius et al., 2018).
The topology of a biological network already containts lots of information. For example, biological network analysis with graph neural networks (GNNs) (Muzio et al., 2020) has been used to predict novel disease-relevant protein-protein and ligand-protein (i.e., drug-target) interactions (Li and Gao, 2019; Zhang et al., 2021). Network inference and graph representations can also be exploited to learn and predict across different types of -omics data (e.g., learn from bulk RNA-seq and predict single-cell RNA-seq) (Hasibi and Michoel, 2021). Molecular interpretation of single-cell RNA-seq data has been also obtained using knowledge-primed neural networks by matching biological and neural network topology (Fortelny and Bock, 2020).
Biochemical profiling of biological fluids (e.g., serum, urine, sweat) as well as tissue samples/extracts can be employed to determine the extracellular and intracellular pools that serve as initial conditions of the model. In particular, the steady-state concentrations of chemical species (e.g., small molecules) can be estimated using experimental metabolomics, whereas proteomics can be used to confirm gene expression patterns, since changes in mRNA levels often correlate poorly with changes in protein expression (Weston and Hood, 2004). Recently, ML has been applied to chromatographic techniques (e.g., liquid or gas chromatography) coupled to mass spectrometry for the identification of specific (e.g., disease-associated) metabolic pathways (Pirhaji et al., 2016; Toubiana et al., 2019). Processing of proteomic and metabolomic datasets have also been used in combination with systems biology and deep learning to discriminate among different disease phenotypes (Kopylov et al., 2021). Using ML, metabolite concentrations have been even predicted from proteomic data (Zelezniak et al., 2018) or from network topology-based optimizations (Tepper et al., 2013; Küken et al., 2019). Protein abundance can be also estimated by integrating network models with transcriptomics (Li et al., 2022) or metabolomic (Di Filippo et al., 2022) data.
Experimental fluxomics data can be used to estimate substrate flow rates within the biological networks. Steady-state and isotopically non-stationary flux analysis is typically based on labeling experiments, like metabolite isotopic enrichment studies that employ 13C-NMR spectroscopy (Winter and Krömer, 2013; Niedenführ et al., 2015). In the case of metabolic networks, flux analysis requires the knowledge of the labeling patterns (i.e., atomic transitions) of each chemical reactions. Since fluxomics data is generally very difficult to obtain, predicting fluxes using ML is highly advised. As an illustration, protein levels have recently been used for training artificial neural networks in order to predict glycolytic reaction fluxes (Ajjolli Nagaraja et al., 2019).
Open issues in model construction include the quality of the reconstructed biological networks (eventually including epigenomic data), because many genes have an unknown function and the literature is often ambiguous or incomplete. In this regard, AI can be used to support/validate the prediction of gene network topology starting from a desired biological function (Shen et al., 2021). Another outstanding challenge is the development of automatic and reliable computational methods for the prediction of the baseline values of concentrations and fluxes, something that as discussed necessarily goes along with the improvement of experimental high-throughput methods and availability of transparent high-quality multi-omics data.
3.3 Model simulation
The successful parametrization of biological network models with AI and multi-omics data allows the implementation of ODE-based simulation for the phenotype of interest, thereby revealing mechanisms and pathways that are likely to contribute characterizing drug actions as well as systemwide side-effects (Michelson et al., 2006; Prokop and Michelson, 2012). Model parametrization is a daunting task, but fortunately biological systems do obey multiple constraints, from simple ones (e.g., flux and mass balance) to more complex ones (e.g., genomic stability). For instance, metabolic network models are inherently constrained by reaction stoichiometry and thermodynamics (e.g., Gibbs free energy and Haldane equations) (Chen et al., 2016; Kiparissides and Hatzimanikatis, 2017). Biological constraints make the corresponding high-dimensional mathematical problems amenable to physics-informed neural networks (PINNs) (Raissi et al., 2019; Karniadakis et al., 2021). Embedding physics into ML is particularly relevant for incorporating incomplete or noisy data in complex models and obtain reliable predictions that can be used to e.g., solve prohibitively expensive inverse problems such as target deconvolution. Accordingly, model parameter estimation has been enhanced by incorporating the system of ODEs into deep fully connected neural networks (Yazdani et al., 2020; Daneker et al., 2022). In particular, loss functions for ML methods can be formulated on the basis of specific simulation outputs that depend on model equations. Notably, systems biology-informed AI can be used to select among different reaction kinetic schemes (e.g., mass-balance, S-systems, log-linear/lin-log, Michaelis-Menten, generalized Hill, convenience, modular) according to the available mechanistic knowledge, experimental datasets, and desired level of details (Du et al., 2016; Kim et al., 2018). Several approaches combining deep learning with GNNs or GANNs have been recently used to fully parametrize ODE-based genome-scale metabolic models in terms of Michaelis-Menten affinity constants (Choudhury et al., 2022) or enzyme turnover numbers (Li et al., 2021), respectively. Bayesian meta-modeling is another systems biology-based approach that uses different mathematical representations, scales, and levels of granularity from prior models in order to simulate cell activity (Raveh et al., 2021). In addition, using ML some aspects of the system’s dynamics can be learned from time-series data whenever these are available (Costello and Martin, 2018).
Due to the high cost and difficulty underlying the experimental determination of kinetic information, ODE-based systems are often underdetermined (e.g., the number of variables is greater than the number of known parameters). The increasing availability of useful curated data notwithstanding, at the moment the sole viable solution to the underdetermination problem is to step up the number of different parametrizations of the same system. Statistical approaches [e.g., (Liepe et al., 2014; Valderrama-Bahamóndez and Fröhlich, 2019)] can be used to obtain good coverage of the parameters space in independent realizations that can thus be numerically solved in parallel (i.e., using ensemble solving). Massively parallel resolution is also beneficial for parameter estimation procedures as well as for neural networks training in surrogate modeling (Renardy et al., 2018; Engel et al., 2019; Vanhaelen, 2022). Recent developments in graphics processing unit (GPU)-accelerated ODE solvers allows investigating large-scale models of complex biological systems at reduced computational costs [for example, see (Tangherloni et al. (2017)].
Overall, stoichiometric, thermodynamic, and kinetic information can be extracted from multi-omics data (Caudai et al., 2021) and processed to enable the generation of ODE-based genome-scale biological network models, a methodology that so far has been limited to simple organisms for applications in metabolic engineering and synthetic biology (Karr et al., 2012; Chakrabarti et al., 2013; Almquist et al., 2014; Srinivasan et al., 2015; Miskovic et al., 2019; St. John et al., 2019). However, modeling and simulation of human cells and tissues promises to enhance our understanding of therapeutical and/or toxicological endpoints in drug discovery by quantitatively uncovering the behavior of many molecular species (proteins, enzymes, receptors, small molecules) that we normally do not have access to in experimental settings. The critical contribution of AI to biological modeling and simulation is the adaptive nature of the algorithms to newly generated data and knowledge, which dynamically affects the probability of performing certain operations rather than others. Embedding biological principles into AI-based approaches will strongly facilitate the computational manipulation of biological networks and the association of phenotypes to targets (i.e., target deconvolution). In silico drug target deconvolution based on modeling and simulation of genome-scale biological networks is a large combinatorial problem. Deep reinforcement learning (DRL) is a very powerful AI technique that presents unique opportunities to map the high-dimensional states underlying the dynamic behavior of ODE-based systems onto appropriate physics-informed actions, e.g., the allowed set of perturbations that result in the most favourable outcome in the long term (Pawar and Maulik, 2021).
4 Conclusion and outlook
The integration of the predictive power of AI (statistical) with the mechanistic understanding of modeling and simulation (deterministic) is expected to give a substantial contribution to the pharmaceutical industry. In particular, convergence of data-driven and theoretical approaches is an important step to complete the data-model-data cycle that is necessary to solve the problem of parameter estimation and elucidate biological system structure, mechanisms, and dynamics (Wang et al., 2018). AI-enabled modeling and simulation are promising tools to improve data interpretation and distinguish between changes that are caused by a disease from those that cause the disease. Next-generation systems biology will undoubtedly benefit from AI methods capable of converting multi-omics data at different scales into actionable knowledge (Nielsen, 2017; Angione, 2019), especially considering the expected advances in data collection from patients (e.g., biosensors for measuring the concentration of chemical species from body fluids) (Jin et al., 2020; Bhave et al., 2021; Phatak et al., 2021). In turn, personalized datasets are poised to substantially enhance the ability of AI for parametrizing quantitative systems pharmacology (QSP) models that combine systems biology with pharmacokinetics and pharmacodynamics (PK/PD) in order to find optimized therapeutics for individual patients or populations with a given disease [for example, see (McEwen et al., 2021)]. Put in perspective, patient-specific biological network models will form the basis for the construction of medical digital twins (or “virtual patients”) [(An and Cockrell, 2022) and references therein]. Genomic data, such as single nucleotide polymorphisms (SNPs), can be incorporated into genome-scale biological network models in terms of enzyme/receptor kinetic parameters (e.g., protein expression/abundance, substrate affinity, reaction or pathway velocity) (Yamada et al., 2001; Jamshidi et al., 2002; Jamshidi and Palsson, 2006). Accordingly, systems biology and ML can be used (for instance, within genome-wide association studies) to discover SNP-disease association and possibly map individual genetic variations to personalized models (Reilly et al., 2013; Ho et al., 2019; Tong et al., 2021b; Mieth et al., 2021; Foguet et al., 2022).
Overall, the AI-enabled computational framework for the analysis of biological networks is anticipated to play a major role in improving drug target identification and validation, qualify any potentially associated side-effect, identify efficacy and toxicity biomarkers, and help with hypothesis generation, optimal experimental design, as well as testing for disease understanding and identification of disease biomarkers (risk, presence, or treatment selection) (Mardinoglu et al., 2014; Zielinski et al., 2015; Raškevičius et al., 2018; Gu et al., 2019; Turanli et al., 2019; Bintener et al., 2020; Gatto et al., 2020; Proffitt et al., 2022). All these aspects are necessary to support the decision-making process of pharmaceutical companies towards de-risking and ultimately accelerating the time-to-market of new effective and safe medicines for patients.
Statements
Author contributions
The author confirms being the sole contributor of this work and has approved it for publication.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
MD was employed by Netabolics SRL.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1
AbbasiM.SantosB. P.PereiraT. C.SofiaR.MonteiroN. R. C.SimõesC. J. V.et al (2022). Designing optimized drug candidates with generative adversarial network. J. Cheminform14, 40. 10.1186/s13321-022-00623-6
2
AbdolhosseiniF.AzarkhaliliB.MaazallahiA.KamalA.MotahariS. A.Sharifi-ZarchiA.et al (2019). Cell identity codes: Understanding cell identity from gene expression profiles using deep neural networks. Sci. Rep.9, 2342. 10.1038/s41598-019-38798-y
3
Ajjolli NagarajaA.FontaineN.DelsautM.ChartonP.DamourC.OffmannB.et al (2019). Flux prediction using artificial neural network (ANN) for the upper part of glycolysis. PLoS One14, e0216178. 10.1371/journal.pone.0216178
4
AlmquistJ.CvijovicM.HatzimanikatisV.NielsenJ.JirstrandM. (2014). Kinetic models in industrial biotechnology - improving cell factory performance. Metab. Eng.24, 38–60. 10.1016/j.ymben.2014.03.007
5
AnG.CockrellC. (2022). Drug development digital twins for drug discovery, testing and repurposing: A schema for requirements and development. Front. Syst. Biol.2. 10.3389/fsysb.2022.928387
6
AngioneC. (2019). Human systems biology and metabolic modelling: A review-from disease metabolism to precision medicine. Biomed. Res. Int.2019, 8304260. 10.1155/2019/8304260
7
ArrellD. K.TerzicA. (2010). Network systems biology for drug discovery. Clin. Pharmacol. Ther.88, 120–125. 10.1038/clpt.2010.91
8
AttarwalaH. (2010). TGN1412: From discovery to disaster. J. Young Pharm.2, 332–336. 10.4103/0975-1483.66810
9
BensonN. (2015). Network-based discovery through mechanistic systems biology. Implications for applications - SMEs and drug discovery: Where the action is. Drug Discov. Today Technol.15, 41–48. 10.1016/j.ddtec.2015.07.003
10
BhaveG.ChenJ. C.SingerA.SharmaA.RobinsonJ. T. (2021). Distributed sensor and actuator networks for closed-loop bioelectronic medicine. Mater. Today46, 125–135. 10.1016/j.mattod.2020.12.020
11
BintenerT.PachecoM. P.SauterT. (2020). Towards the routine use of in silico screenings for drug discovery using metabolic modelling. Biochem. Soc. Trans.48, 955–969. 10.1042/bst20190867
12
BlanchardA. E.StanleyC.BhowmikD. (2021). Using GANs with adaptive training data to search for new molecules. J. Cheminform13, 14. 10.1186/s13321-021-00494-3
13
BoothB.GlassmanR.MaP. (2003). Oncology's trials. Nat. Rev. Drug Discov.2, 609–610. 10.1038/nrd1158
14
BrunkE.SahooS.ZielinskiD. C.AltunkayaA.DrägerA.MihN.et al (2018). Recon3D enables a three-dimensional view of gene variation in human metabolism. Nat. Biotechnol.36, 272–281. 10.1038/nbt.4072
15
CamachoD. M.CollinsK. M.PowersR. K.CostelloJ. C.CollinsJ. J. (2018). Next-generation machine learning for biological networks. Cell173, 1581–1592. 10.1016/j.cell.2018.05.015
16
CaudaiC.GaliziaA.GeraciF.Le PeraL.MoreaV.SalernoE.et al (2021). AI applications in functional genomics. Comput. Struct. Biotechnol. J.19, 5762–5790. 10.1016/j.csbj.2021.10.009
17
ChakrabartiA.MiskovicL.SohK. C.HatzimanikatisV. (2013). Towards kinetic modeling of genome-scale metabolic networks without sacrificing stoichiometric, thermodynamic and physiological constraints. Biotechnol. J.8, 1043–1057. 10.1002/biot.201300091
18
ChavaliA. K.RheeS. Y. (2017). Bioinformatics tools for the identification of gene clusters that biosynthesize specialized metabolites. Briefings Bioinforma.19, 1022–1034. 10.1093/bib/bbx020
19
ChenY. C.YuanR. S.AoP.XuM. J.ZhuX. M. (2016). Towards stable kinetics of large metabolic networks: Nonequilibrium potential function approach. Phys. Rev. E93, 062409. 10.1103/PhysRevE.93.062409
20
ChoudhuryS.MoretM.SalvyP.WeilandtD.HatzimanikatisV.MiskovicL. (2022). Reconstructing kinetic models for dynamical studies of metabolism using generative adversarial networks. bioRxiv2022, 475020. 10.1038/s42256-022-00519-y
21
CostelloZ.MartinH. G. (2018). A machine learning approach to predict metabolic pathway dynamics from time-series multiomics data. npj Syst. Biol. Appl.4, 19. 10.1038/s41540-018-0054-3
22
Cuperlovic-CulfM. (2018). Machine learning methods for analysis of metabolic data and metabolic pathway modeling. Metabolites8, 4. 10.3390/metabo8010004
23
DanekerM.ZhangZ.KarniadakisG. E.LuL. (2022). Systems Biology: Identifiability analysis and parameter identification via systems-biology informed neural networks, 1–18. arXiv.
24
Di FilippoM.PesciniD.GaluzziB. G.BonanomiM.GaglioD.ManganoE.et al (2022). Integrate: Model-based multi-omics data integration to characterize multi-level metabolic regulation. PLoS Comput. Biol.18, e1009337. 10.1371/journal.pcbi.1009337
25
DíazÓ.DaltonJ. A. R.GiraldoJ. (2019). Artificial intelligence: A novel approach for drug discovery. Trends Pharmacol. Sci.40, 550–551. 10.1016/j.tips.2019.06.005
26
DowdenH.MunroJ. (2019). Trends in clinical success rates and therapeutic focus. Nat. Rev. Drug Discov.18, 495–496. 10.1038/d41573-019-00074-z
27
DuB.ZielinskiD. C.KavvasE. S.DrägerA.TanJ.ZhangZ.et al (2016). Evaluation of rate law approximations in bottom-up kinetic models of metabolism. BMC Syst. Biol.10, 40. 10.1186/s12918-016-0283-2
28
EngelG. (2019). “Neural networks to approximate solutions of ordinary differential equations,” in Intelligent computing. Editors AraiK.BhatiaR.KapoorS. (Cham: Springer International Publishing), 776–784. 10.1007/978-3-030-22871-2_54
29
FlorezJ. C. (2017). Mining the genome for therapeutic targets. Diabetes66, 1770–1778. 10.2337/dbi16-0069
30
FoguetC.XuY.RitchieS. C.LambertS. A.PersynE.NathA. P.et al (2022). Genetically personalised organ-specific metabolic models in health and disease. 10.1101/2022.03.25.22272958v1
31
FortelnyN.BockC. (2020). Knowledge-primed neural networks enable biologically interpretable deep learning on single-cell sequencing data. Genome Biol.21, 190. 10.1186/s13059-020-02100-5
32
GalatiS.Di StefanoM.MartinelliE.MacchiaM.MartinelliA.PoliG.et al (2022). VenomPred: A machine learning based platform for molecular toxicity predictions. Int. J. Mol. Sci.23. 10.3390/ijms23042105
33
GattoF.FerreiraR.NielsenJ. (2020). Pan-cancer analysis of the metabolic reaction network. Metab. Eng.57, 51–62. 10.1016/j.ymben.2019.09.006
34
GaudeletT.DayB.JamasbA. R.SomanJ.RegepC.LiuG.et al (2021). Utilizing graph machine learning within drug discovery and development. Brief. Bioinform22. 10.1093/bib/bbab159
35
GhorbaniM.KarimiH. (2015). Bioinformatics approaches for gene finding. Int. J. Sci. Res. Sci. Technol.4, 12–15.
36
GoldbeterA. (2018). Dissipative structures in biological systems: Bistability, oscillations, spatial patterns and waves. Philos. Trans. A Math. Phys. Eng. Sci.376. 10.1098/rsta.2017.0376
37
GoldbeterA.YanJ. (2022). Multi-synchronization and other patterns of multi-rhythmicity in oscillatory biological systems. Interface Focus12, 20210089. 10.1098/rsfs.2021.0089
38
GöllerA. H.KuhnkeL.ter LaakA.MeierK.HillischA. (2022). Machine learning applied to the modeling of pharmacological and ADMET endpoints. Methods Mol. Biol.2390, 61–101. 10.1007/978-1-0716-1787-8_2
39
GuC.KimG. B.KimW. J.KimH. U.LeeS. Y. (2019). Current status and applications of genome-scale metabolic models. Genome Biol.20, 121. 10.1186/s13059-019-1730-3
40
HarrisonR. K. (2016). Phase II and phase III failures: 2013-2015. Nat. Rev. Drug Discov.15, 817–818. 10.1038/nrd.2016.184
41
HasibiR.MichoelT. (2021). A Graph Feature Auto-Encoder for the prediction of unobserved node features on biological networks. BMC Bioinforma.22, 525. 10.1186/s12859-021-04447-3
42
HayM.ThomasD. W.CraigheadJ. L.EconomidesC.RosenthalJ. (2014). Clinical development success rates for investigational drugs. Nat. Biotechnol.32, 40–51. 10.1038/nbt.2786
43
HoD. S. W.SchierdingW.WakeM.SafferyR.O’SullivanJ. (2019). Machine learning SNP based prediction for precision medicine. Front. Genet.10, 267. 10.3389/fgene.2019.00267
44
HughesR. E.ElliottR. J. R.DawsonJ. C.CarragherN. O. (2021). High-content phenotypic and pathway profiling to advance drug discovery in diseases of unmet need. Cell Chem. Biol.28, 338–355. 10.1016/j.chembiol.2021.02.015
45
IdekerT.GalitskiT.HoodL. (2001). A new approach to decoding life: Systems biology. Annu. Rev. Genom. Hum. Genet.2, 343–372. 10.1146/annurev.genom.2.1.343
46
InstituteT. (1995). “Of medicine committee to review the fialuridine clinical,” in Review of the fialuridine (FIAU) clinical trials. Editors ManningF. J.SwartzM. (Washington (DC): National Academies Press (US). Copyright 1995 by the National Academy of Sciences. All rights reserved.
47
IQVIA (2021). The global use of medicines 2022 and outlook to 2026. Durham, NC: Institute Report.
48
IwataH.KojimaR.OkunoY. (2020). An in silico approach for integrating phenotypic and target-based approaches in drug discovery. Mol. Inf.39, e1900096. 10.1002/minf.201900096
49
JamshidiN.PalssonB. Ø. (2006). Systems biology of SNPs. Mol. Syst. Biol.2, 38. 10.1038/msb4100077
50
JamshidiN.WibackS. J.PalssonB. B. (2002). In silico model-driven assessment of the effects of single nucleotide polymorphisms (SNPs) on human red blood cell metabolism. Genome Res.12, 1687–1692. 10.1101/gr.329302
51
JiaL.GaoH. (2022). Machine learning for in silico ADMET prediction. Methods Mol. Biol.2390, 447–460. 10.1007/978-1-0716-1787-8_20
52
Jiménez-LunaJ.GrisoniF.SchneiderG. (2020). Drug discovery with explainable artificial intelligence. Nat. Mach. Intell.2, 573–584. 10.1038/s42256-020-00236-4
53
Jiménez-LunaJ.GrisoniF.WeskampN.SchneiderG. (2021). Artificial intelligence in drug discovery: Recent advances and future perspectives. Expert Opin. Drug Discov.16, 949–959. 10.1080/17460441.2021.1909567
54
JinX.LiuC.XuT.SuL.ZhangX. (2020). Artificial intelligence biosensors: Challenges and prospects. Biosens. Bioelectron.165, 112412. 10.1016/j.bios.2020.112412
55
KarniadakisG. E.KevrekidisI. G.LuL.PerdikarisP.WangS.YangL. (2021). Physics-informed machine learning. Nat. Rev. Phys.3, 422–440. 10.1038/s42254-021-00314-5
56
KarpP. D.WeaverD.LatendresseM. (2018). How accurate is automated gap filling of metabolic models?BMC Syst. Biol.12, 73. 10.1186/s12918-018-0593-7
57
KarrJ. R.SanghviJ. C.MacklinD. N.GutschowM. V.JacobsJ. M.BolivalB.Jr.et al (2012). A whole-cell computational model predicts phenotype from genotype. Cell150, 389–401. 10.1016/j.cell.2012.05.044
58
KellD. B.GoodacreR. (2014). Metabolomics and systems pharmacology: Why and how to model the human metabolic network for drug discovery. Drug Discov. Today19, 171–182. 10.1016/j.drudis.2013.07.014
59
KimO. D.RochaM.MaiaP. (2018). A review of dynamic modeling approaches and their application in computational strain optimization for metabolic engineering. Front. Microbiol.9, 1690. 10.3389/fmicb.2018.01690
60
KiparissidesA.HatzimanikatisV. (2017). Thermodynamics-based metabolite sensitivity analysis in metabolic networks. Metab. Eng.39, 117–127. 10.1016/j.ymben.2016.11.006
61
KolaI.LandisJ. (2004). Can the pharmaceutical industry reduce attrition rates?Nat. Rev. Drug Discov.3, 711–716. 10.1038/nrd1470
62
KopylovA. T.PetrovskyD. V.StepanovA. A.RudnevV. R.MalsagovaK. A.ButkovaT. V.et al (2021). Convolutional neural network in proteomics and metabolomics for determination of comorbidity between cancer and schizophrenia. J. Biomed. Inf.122, 103890. 10.1016/j.jbi.2021.103890
63
KükenA.Eloundou-MbebiJ. M. O.BaslerG.NikoloskiZ. (2019). Cellular determinants of metabolite concentration ranges. PLoS Comput. Biol.15, e1006687. 10.1371/journal.pcbi.1006687
64
LähnemannD.KösterJ.SzczurekE.McCarthyD. J.HicksS. C.RobinsonM. D.et al (2020). Eleven grand challenges in single-cell data science. Genome Biol.21, 31. 10.1186/s13059-020-1926-6
65
LeonD.MarkelS. (2006). Silico technologies in drug target identification and validation. Boca Raton: CRC Press.
66
LiD.GaoJ. (2019). Towards perturbation prediction of biological networks using deep learning. Sci. Rep.9, 11941. 10.1038/s41598-019-48391-y
67
LiF.YuanL.LuH.LiG.ChenY.EngqvistM. K. M.et al (2021). Deep learning based <em>k</em>cat prediction enables improved enzyme constrained model reconstruction. bioRxiv2021, 455417.
68
LiG. H.HanF.XiaoF. H.GuK. S.ShenQ.XuW.et al (2022). System-level metabolic modeling facilitates unveiling metabolic signature in exceptional longevity. Aging Cell21, e13595. 10.1111/acel.13595
69
LiY.WangS.UmarovR.XieB.FanM.LiL.et al (2018). DEEPre: Sequence-based enzyme EC number prediction by deep learning. Bioinformatics34, 760–769. 10.1093/bioinformatics/btx680
70
LiepeJ.KirkP.FilippiS.ToniT.BarnesC. P.StumpfM. P. H. (2014). A framework for parameter estimation and model selection from experimental data in systems biology using approximate Bayesian computation. Nat. Protoc.9, 439–456. 10.1038/nprot.2014.025
71
MamadaH.NomuraY.UesawaY. (2022). Novel QSAR approach for a regression model of clearance that combines DeepSnap-deep learning and conventional machine learning. ACS Omega7, 17055–17062. 10.1021/acsomega.2c00261
72
MardinogluA.AgrenR.KampfC.AsplundA.UhlenM.NielsenJ. (2014). Genome-scale metabolic modelling of hepatocytes reveals serine deficiency in patients with non-alcoholic fatty liver disease. Nat. Commun.5, 3083. 10.1038/ncomms4083
73
MayE.ShahS.TaylorK.MirandaW.CruzM. J. (2022). Measuring the return from pharmaceutical innovation: Deloitte Insights.
74
McEwenS. C.MerrillD. A.BramenJ.PorterV.PanosS.KaiserS.et al (2021). A systems-biology clinical trial of a personalized multimodal lifestyle intervention for early Alzheimer's disease. Alzheimers Dement. (N Y)7, e12191. 10.1002/trc2.12191
75
MichelsonS.ScherrerD.BangsA. L. (2006). “Target identification and validation using human simulation models,” in In silico technologies in drug target identification and validation. Editors SchneiderR.VossH. (Boca Raton, FL: CRC Press, Taylor and Francis Group, LLC), 345–375.
76
MiethB.RozierA.RodriguezJ. A.HöhneM. M. C.GörnitzN.MüllerK.-R. (2021). DeepCOMBI: Explainable artificial intelligence for the analysis and discovery in genome-wide association studies. NAR Genomics Bioinforma.3. 10.1093/nargab/lqab065
77
MiskovicL.BéalJ.MoretM.HatzimanikatisV. (2019). Uncertainty reduction in biochemical kinetic models: Enforcing desired model properties. PLoS Comput. Biol.15, e1007242. 10.1371/journal.pcbi.1007242
78
MoffatJ. G.VincentF.LeeJ. A.EderJ.PrunottoM. (2017). Opportunities and challenges in phenotypic drug discovery: An industry perspective. Nat. Rev. Drug Discov.16, 531–543. 10.1038/nrd.2017.111
79
MuzioG.O’BrayL.BorgwardtK. (2020). Biological network analysis with deep learning. Briefings Bioinforma.22, 1515–1530. 10.1093/bib/bbaa257
80
NakaiK.HortonP. (2007). “Computational prediction of subcellular localization,” in Protein targeting protocols. Editor van der GiezenM. (Totowa, NJ: Humana Press), 429–466. 10.1007/978-1-59745-466-7_29
81
NiedenführS.WiechertW.NöhK. (2015). How to measure metabolic fluxes: A taxonomic guide for 13 C fluxomics. Curr. Opin. Biotechnol.34, 82–90. 10.1016/j.copbio.2014.12.003
82
NielsenJ. (2017). Systems biology of metabolism: A driver for developing personalized and precision medicine. Cell Metab.25, 572–579. 10.1016/j.cmet.2017.02.002
83
PanS.ReedJ. L. (2018). Advances in gap-filling genome-scale metabolic models and model-driven experiments lead to novel metabolic discoveries. Curr. Opin. Biotechnol.51, 103–108. 10.1016/j.copbio.2017.12.012
84
PaulD.SanapG.ShenoyS.KalyaneD.KaliaK.TekadeR. K. (2021). Artificial intelligence in drug discovery and development. Drug Discov. Today26, 80–93. 10.1016/j.drudis.2020.10.010
85
PawarS.MaulikR. (2021). Distributed deep reinforcement learning for simulation control. Mach. Learn. Sci. Technol.2, 025029. 10.1088/2632-2153/abdaf8
86
PeckR. W.HolsteinS. A.GraafP. H. (2022). Bia 10‐2474: Some lessons are clear but important questions remain unanswered. Clin Pharma Ther.111, 343–345. 10.1002/cpt.2495
87
PhatakA. A.WielandF.-G.VempalaK.VolkmarF.MemmertD. (2021). Artificial intelligence based body sensor network framework-narrative review: Proposing an end-to-end framework using wearable sensors, real-time location systems and artificial intelligence/machine learning algorithms for data collection, data mining and knowledge discovery in sports and healthcare. Sports Med. - Open7, 79. 10.1186/s40798-021-00372-0
88
PirhajiL.MilaniP.LeidlM.CurranT.Avila-PachecoJ.ClishC. B.et al (2016). Revealing disease-associated pathways by network integration of untargeted metabolomics. Nat. Methods13, 770–776. 10.1038/nmeth.3940
89
PritchardJ. F.Jurima-RometM.ReimerM. L.MortimerE.RolfeB.CayenM. N. (2003). Making better drugs: Decision gates in non-clinical drug development. Nat. Rev. Drug Discov.2, 542–553. 10.1038/nrd1131
90
ProffittC.BidkhoriG.LeeS.TebaniA.MardinogluA.UhlenM.et al (2022). Genome-scale metabolic modelling of the human gut microbiome reveals changes in the glyoxylate and dicarboxylate metabolism in metabolic disorders. iScience25, 104513. 10.1016/j.isci.2022.104513
91
ProkopA.MichelsonS. (2012). Systems biology in biotech & pharma. Dordrecht: Springer.
92
PrykhodkoO.JohanssonS. V.KotsiasP.-C.Arús-PousJ.BjerrumE. J.EngkvistO.et al (2019). A de novo molecular generation method using latent vector based generative adversarial network. J. Cheminform11, 74. 10.1186/s13321-019-0397-9
93
PujolA.MoscaR.FarrésJ.AloyP. (2010). Unveiling the role of network and systems biology in drug discovery. Trends Pharmacol. Sci.31, 115–123. 10.1016/j.tips.2009.11.006
94
RaiS.RajU.VaradwajP. K. (2018). Systems biology: A powerful tool for drug development. Ctmc18, 1745–1754. 10.2174/1568026618666181025113226
95
RaissiM.PerdikarisP.KarniadakisG. E. (2019). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J. Comput. Phys.378, 686–707. 10.1016/j.jcp.2018.10.045
96
RaškevičiusV.MikalayevaV.AntanavičiūtėI.CeslevičienėI.SkeberdisV. A.KairysV.et al (2018). Genome scale metabolic models as tools for drug design and personalized medicine. PLoS One13, e0190636.
97
RavehB.SunL.WhiteK. L.SanyalT.TempkinJ.ZhengD.et al (2021). Bayesian metamodeling of complex biological systems across varying representations. Proc. Natl. Acad. Sci. U. S. A.118. 10.1073/pnas.2104559118
98
ReedJ. L.FamiliI.ThieleI.PalssonB. O. (2006). Towards multidimensional genome annotation. Nat. Rev. Genet.7, 130–141. 10.1038/nrg1769
99
ReillyD.HaoK.JensenM. K.GirmanC. J.RimmE. B. (2013). Use of systems biology approaches to analysis of genome-wide association studies of myocardial infarction and blood cholesterol in the nurses' health study and health professionals' follow-up study. PLOS ONE8, e85369. 10.1371/journal.pone.0085369
100
RenardyM.YiT.-M.XiuD.ChouC.-S. (2018). Parameter uncertainty quantification using surrogate models applied to a spatial model of yeast mating polarization. PLoS Comput. Biol.14, e1006181. 10.1371/journal.pcbi.1006181
101
RobertsR. A. (2018). Understanding drug targets: No such thing as bad news. Drug Discov. Today23, 1925–1928. 10.1016/j.drudis.2018.05.028
102
SaliA. (2021). From integrative structural biology to cell biology. J. Biol. Chem.296, 100743. 10.1016/j.jbc.2021.100743
103
SchneiderG.FechnerU. (2004). Advances in the prediction of protein targeting signals. Proteomics4, 1571–1580. 10.1002/pmic.200300786
104
ShenJ.LiuF.TuY.TangC. (2021). Finding gene network topologies for given biological function with recurrent neural network. Nat. Commun.12, 3125. 10.1038/s41467-021-23420-5
105
SinglaJ.WhiteK. L. (2021). A community approach to whole-cell modeling. Curr. Opin. Syst. Biol.26, 33–38. 10.1016/j.coisb.2021.03.009
106
SrinivasanS.CluettW. R.MahadevanR. (2015). Constructing kinetic models of metabolism at genome-scales: A review. Biotechnol. J.10, 1345–1359. 10.1002/biot.201400522
107
SteedmanM.TaylorK. (2020). Measuring the return from pharmaceutical innovation. London, United Kingdom: Deloitte Insights.
108
St. JohnP. C.StrutzJ.BroadbeltL. J.TyoK. E. J.BombleY. J. (2019). Bayesian inference of metabolic kinetics from genome-scale multiomics data. PLoS Comput. Biol.15, e1007424. 10.1371/journal.pcbi.1007424
109
TakebeT.ImaiR.OnoS. (2018). The current status of drug discovery and development as originated in United States academia: The influence of industrial and academic collaboration on drug discovery and development. Clin. Transl. Sci.11, 597–606. 10.1111/cts.12577
110
TangherloniA.NobileM. S.BesozziD.MauriG.CazzanigaP. (2017). Lassie: Simulating large-scale models of biochemical systems on GPUs. BMC Bioinforma.18, 246. 10.1186/s12859-017-1666-0
111
TepperN.NoorE.Amador-NoguezD.HaraldsdóttirH. S.MiloR.RabinowitzJ.et al (2013). Steady-state metabolite concentrations reflect a balance between maximizing enzyme efficiency and minimizing total metabolite load. PLoS One8, e75370. 10.1371/journal.pone.0075370
112
ThieleI.SahooS.HeinkenA.HertelJ.HeirendtL.AurichM. K.et al (2020). Personalized whole-body models integrate metabolism, physiology, and the gut microbiome. Mol. Syst. Biol.16, e8982. 10.15252/msb.20198982
113
TongH.KükenA.Razaghi-MoghadamZ.NikoloskiZ. (2021). Characterization of effects of genetic variants via genome-scale metabolic modelling. Cell. Mol. Life Sci.78, 5123–5138. 10.1007/s00018-021-03844-4
114
TongX.LiuX.TanX.LiX.JiangJ.XiongZ.et al (2021). Generative models for de novo drug design. J. Med. Chem.64, 14011–14027. 10.1021/acs.jmedchem.1c00927
115
ToubianaD.PuzisR.WenL.SikronN.KurmanbayevaA.SoltabayevaA.et al (2019). Combined network analysis and machine learning allows the prediction of metabolic pathways from tomato metabolomics data. Commun. Biol.2, 214. 10.1038/s42003-019-0440-4
116
TuranliB.ZhangC.KimW.BenfeitasR.UhlenM.ArgaK. Y.et al (2019). Discovery of therapeutic agents for prostate cancer using genome-scale metabolic modeling and drug repositioning. EBioMedicine42, 386–396. 10.1016/j.ebiom.2019.03.009
117
Valderrama-BahamóndezG. I.FröhlichH. (2019). MCMC techniques for parameter estimation of ODE based models in systems biology. Front. Appl. Math. Statistics5. 10.3389/fams.2019.00055
118
van HasseltJ. G. C.IyengarR. (2019). Systems pharmacology: Defining the interactions of drug combinations. Annu. Rev. Pharmacol. Toxicol.59, 21–40. 10.1146/annurev-pharmtox-010818-021511
119
VanhaelenQ. (2022). Computational methods for estimating the kinetic parameters of biological systems. New York, NY: Humana.
120
VincentF.NuedaA.LeeJ.SchenoneM.PrunottoM.MercolaM. (2022). Phenotypic drug discovery: Recent successes, lessons learned and new directions. Nat. Rev. Drug Discov.10.1038/s41573-022-00472-w
121
WangY.ZhangX.-S.ChenL. (2018). Integrating data- and model-driven strategies in systems biology. BMC Syst. Biol.12, 38. 10.1186/s12918-018-0562-1
122
WangZ.ChenY.LiY. (2004). A brief review of computational gene prediction methods. Genomics, Proteomics Bioinforma.2, 216–221. 10.1016/s1672-0229(04)02028-5
123
WestonA. D.HoodL. (2004). Systems biology, proteomics, and the future of health care: Toward predictive, preventative, and personalized medicine. J. Proteome Res.3, 179–196. 10.1021/pr0499693
124
WinterG.KrömerJ. O. (2013). Fluxomics - connecting 'omics analysis and phenotypes. Environ. Microbiol.15, 1901–1916. 10.1111/1462-2920.12064
125
YamadaK.ChenZ.RozenR.MatthewsR. G. (2001). Effects of common polymorphisms on the properties of recombinant human methylenetetrahydrofolate reductase. Proc. Natl. Acad. Sci. U.S.A.98, 14853–14858. 10.1073/pnas.261469998
126
YazdaniA.LuL.RaissiM.KarniadakisG. E. (2020). Systems biology informed deep learning for inferring parameters and hidden dynamics. PLoS Comput. Biol.16, e1007575. 10.1371/journal.pcbi.1007575
127
ZampieriG.VijayakumarS.YaneskeE.AngioneC. (2019). Machine and deep learning meet genome-scale metabolic modeling. PLoS Comput. Biol.15, e1007084. 10.1371/journal.pcbi.1007084
128
ZelezniakA.VowinckelJ.CapuanoF.MessnerC. B.DemichevV.PolowskyN.et al (2018). Machine learning predicts the yeast metabolome from the quantitative proteome of kinase knockouts. Cell Syst.7, 269–283. e6. 10.1016/j.cels.2018.08.001
129
ZhangX.-M.LiangL.LiuL.TangM.-J. (2021). Graph neural networks and their current applications in bioinformatics. Front. Genet.12. 10.3389/fgene.2021.690049
130
ZielinskiD. C.FilippF. V.BordbarA.JensenK.SmithJ. W.HerrgardM. J.et al (2015). Pharmacogenomic and clinical data link non-pharmacokinetic metabolic dysregulation to drug side effect pathogenesis. Nat. Commun.6, 7101. 10.1038/ncomms8101
Summary
Keywords
modeling and simulation, biological networks, systems biology, drug discovery, artificial intelligence
Citation
DiNuzzo M (2022) How artificial intelligence enables modeling and simulation of biological networks to accelerate drug discovery. Front. Drug. Discov. 2:1019706. doi: 10.3389/fddsv.2022.1019706
Received
15 August 2022
Accepted
14 September 2022
Published
04 October 2022
Volume
2 - 2022
Edited by
José L. Medina-Franco, National Autonomous University of Mexico, Mexico
Reviewed by
Abhik Seal, AbbVie, United States
Updates
Copyright
© 2022 DiNuzzo.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Mauro DiNuzzo, mauro@netabolics.ai
This article was submitted to In silico Methods and Artificial Intelligence for Drug Discovery, a section of the journal Frontiers in Drug Discovery
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.