Frontiers in Bioinformatics has organized a series of Research Topics which aim to promote debate or to change the perspective on unsettled research in various areas of the field of bioinformatics. To provide a critical evaluation of the state of research in the field of "Protein bioinformatics", leading experts have been invited by our Specialty Chief Editor - Prof. Daisuke Kihara - to contribute their thoughts and perspectives on current challenges, latest discoveries, recent advances, and future perspectives in the field. This Research Topic solicits the submission of Opinion and Perspective articles that provide analysis related to the field of "Protein bioinformatics" by established researchers in the field. Please note, contributions to this article collection are by invitation only. Please inform the Editorial Office at <a href="mailto:bioinformatics.submissions@frontiersin.org">bioinformatics.submissions@frontiersin.org</a> if you wish to submit to this Research Topic. For more information regarding the article types supported by the journal, please check here: <a href="https://www.frontiersin.org/journals/bioinformatics#article-types">https://www.frontiersin.org/journals/bioinformatics#article-types </a>

Purdue University

Views

Article Views

Article Downloads

Frontiers in Bioinformatics

Protein Bioinformatics

Expert Opinions in Protein bioinformatics: 2022

In 2021, protein bioinformatics underwent a profound and irreversible transformation with the publication of AlphaFold2, a method for predicting protein tertiary structures. AlphaFold2 (<a href="#B1">Jumper et al., 2021</a>) demonstrated significantly higher accuracy compared to existing methods during the 14th Critical Assessment of Techniques for Protein Structure Prediction (CASP14). Notably, its accuracy often rivals that of experimental structure determination techniques like X-ray crystallography, which has proved to be exceptionally valuable in various practical applications.An additional noteworthy aspect of AlphaFold2 is that it represents the first successful end-to-end deep neural network for protein structure prediction. This achievement is particularly remarkable given that protein structure prediction has long been recognized as one of the most challenging problems in bioinformatics and computational biophysics. Unlike existing methods with complex architectures involving intricate steps in a computational modeling pipeline, AlphaFold2 demonstrated that the entire process can be encapsulated within a single neural network.The success of Alphafold2 is resulted from convergence of crucial elements, including the maturation of the protein structure prediction problem, the availability of a substantial dataset, the rapid advancements in deep learning, and the presence of adequate computational resources. This also indicates that the same approach would be possible for other molecular modeling problems, too.Following AlphaFold2, a cascade of related developments ensued, spanning protein docking prediction, algorithms for protein design and drug design, RNA structure prediction, and predictions of the effects of missense mutations, all using deep learning. Consequently, the emergence of AlphaFold2 will be remembered not only as a singular achievement but also as the beginning of the end for the development of general-purpose modeling methods that aim to be applied to any proteins, drugs, RNA, and complexes.In the era of the end of general-purpose modeling problems, an increasing emphasis will be placed on the accurate solutions to individual, fine-grained, and specific tasks. This compilation serves as a testament to the significant research topics that define the post-general-purpose problems era. Gomez et al. explored the applicability of AlphaFold2 models within the context of force spectroscopy experiments (<a href="https://doi.org/10.3389/fbinf.2022.983306">Gomez et al.</a>). Chang et al. delved into the key factors crucial for the successful computational design of peptides, highlighting an important application of protein design methods (<a href="https://doi.org/10.3389/fbinf.2022.1046493">Chang et al.</a>). The article by Neumann et al. the Ben-Hur group specifically addresses the importance of the appropriate selection of negative examples in predicting host-pathogen protein interactions (<a href="https://doi.org/10.3389/fbinf.2022.1083292">Neumann et al.</a>). Lastly, Jeffery provides insights into the various challenges that make protein function prediction a complex task (<a href="https://doi.org/10.3389/fbinf.2023.1222182">Jeffery</a>).As we navigate the landscape of the post-general-purpose method development, the contributions highlighted in this Research Topic reflect the profound impact and continued evolution of methodologies addressing specific molecular tasks. The success of future protein bioinformatics projects will hinge on the seamless integration of a detailed understanding of biological problems and background, coupled with the proper application of machine learning methods, to accomplish specific tasks.DK: Conceptualization, Writing&#x2013;original draft, Writing&#x2013;review and editing.The author(s) declare financial support was received for the research, authorship, and/or publication of this article. DK acknowledges funding from the National Institutes of Health (R01GM133840) and the National Science Foundation (DBI2003635, DBI2146026, IIS2211598, DMS2151678, CMMI1825941, and MCB1925643).The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.The author(s) declared that they were an editorial board member of Frontiers, at the time of submission. This had no impact on the peer review process and the final decision.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.

Editorial: Expert opinions in protein bioinformatics: 2022

In recent years, improvements in protein function prediction methods have led to increased success in annotating protein sequences. However, the functions of over 30% of protein-coding genes remain unknown for many sequenced genomes. Protein functions vary widely, from catalyzing chemical reactions to binding DNA or RNA or forming structures in the cell, and some types of functions are challenging to predict due to the physical features associated with those functions. Other complications in understanding protein functions arise due to the fact that many proteins have more than one function or very small differences in sequence or structure that correspond to different functions. We will discuss some of the recent developments in predicting protein functions and some of the remaining challenges.

In recent years, improvements in protein function prediction methods have led to increased success in annotating protein sequences. However, the functions of over 30% of protein-coding genes remain unknown for many sequenced genomes. Protein functions vary widely, from catalyzing chemical reactions to binding DNA or RNA or forming structures in the cell, and some types of functions are challenging to predict due to the physical features associated with those functions. Other complications in understanding protein functions arise due to the fact that many proteins have more than one function or very small differences in sequence or structure that correspond to different functions. We will discuss some of the recent developments in predicting protein functions and some of the remaining challenges.

Current successes and remaining challenges in protein function prediction

As practitioners of machine learning in the area of bioinformatics we know that the quality of the results crucially depends on the quality of our labeled data. While there is a tendency to focus on the quality of positive examples, the negative examples are equally as important. In this opinion paper we revisit the problem of choosing negative examples for the task of predicting protein-protein interactions, either among proteins of a given species or for host-pathogen interactions and describe important issues that are prevalent in the current literature. The challenge in creating datasets for this task is the noisy nature of the experimentally derived interactions and the lack of information on non-interacting proteins. A standard approach is to choose random pairs of non-interacting proteins as negative examples. Since the interactomes of all species are only partially known, this leads to a very small percentage of false negatives. This is especially true for host-pathogen interactions. To address this perceived issue, some researchers have chosen to select negative examples as pairs of proteins whose sequence similarity to the positive examples is sufficiently low. This clearly reduces the chance for false negatives, but also makes the problem much easier than it really is, leading to over-optimistic accuracy estimates. We demonstrate the effect of this form of bias using a selection of recent protein interaction prediction methods of varying complexity, and urge researchers to pay attention to the details of generating their datasets for potential biases like this.

As practitioners of machine learning in the area of bioinformatics we know that the quality of the results crucially depends on the quality of our labeled data. While there is a tendency to focus on the quality of positive examples, the negative examples are equally as important. In this opinion paper we revisit the problem of choosing negative examples for the task of predicting protein-protein interactions, either among proteins of a given species or for host-pathogen interactions and describe important issues that are prevalent in the current literature. The challenge in creating datasets for this task is the noisy nature of the experimentally derived interactions and the lack of information on non-interacting proteins. A standard approach is to choose random pairs of non-interacting proteins as negative examples. Since the interactomes of all species are only partially known, this leads to a very small percentage of false negatives. This is especially true for host-pathogen interactions. To address this perceived issue, some researchers have chosen to select negative examples as pairs of proteins whose sequence similarity to the positive examples is sufficiently low. This clearly reduces the chance for false negatives, but also makes the problem much easier than it really is, leading to over-optimistic accuracy estimates. We demonstrate the effect of this form of bias using a selection of recent protein interaction prediction methods of varying complexity, and urge researchers to pay attention to the details of generating their datasets for potential biases like this.

On the choice of negative examples for prediction of host-pathogen protein interactions

Peptides are prevalent in biology, mediating as many as 40% of protein-protein interactions, and involved in other cellular functions such as transport and signaling. Their ability to bind with high specificity make them promising therapeutical agents with intermediate properties between small molecules and large biologics. Beyond their biological role, peptides can be programmed to self-assembly, and they are already being used for functions as diverse as oligonuclotide delivery, tissue regeneration or as drugs. However, the transient nature of their interactions has limited the number of structures and knowledge of binding affinities available–and their flexible nature has limited the success of computational pipelines that predict the structures and affinities of these molecules. Fortunately, recent advances in experimental and computational pipelines are creating new opportunities for this field. We are starting to see promising predictions of complex structures, thermodynamic and kinetic properties. We believe in the following years this will lead to robust rational peptide design pipelines with success similar to those applied for small molecule drug discovery.

Peptides are prevalent in biology, mediating as many as 40% of protein-protein interactions, and involved in other cellular functions such as transport and signaling. Their ability to bind with high specificity make them promising therapeutical agents with intermediate properties between small molecules and large biologics. Beyond their biological role, peptides can be programmed to self-assembly, and they are already being used for functions as diverse as oligonuclotide delivery, tissue regeneration or as drugs. However, the transient nature of their interactions has limited the number of structures and knowledge of binding affinities available&#x2013;and their flexible nature has limited the success of computational pipelines that predict the structures and affinities of these molecules. Fortunately, recent advances in experimental and computational pipelines are creating new opportunities for this field. We are starting to see promising predictions of complex structures, thermodynamic and kinetic properties. We believe in the following years this will lead to robust rational peptide design pipelines with success similar to those applied for small molecule drug discovery.

Towards rational computational peptide design

Mechanoactive proteins are essential for a myriad of physiological and pathological processes. Guided by the advances in single-molecule force spectroscopy (SMFS), we have reached a molecular-level understanding of how mechanoactive proteins sense and respond to mechanical forces. However, even SMFS has its limitations, including the lack of detailed structural information during force-loading experiments. That is where molecular dynamics (MD) methods shine, bringing atomistic details with femtosecond time-resolution. However, MD heavily relies on the availability of high-resolution structural data, which is not available for most proteins. For instance, the Protein Data Bank currently has 192K structures deposited, against 231M protein sequences available on Uniprot. But many are betting that this gap might become much smaller soon. Over the past year, the AI-based AlphaFold created a buzz on the structural biology field by being able to predict near-native protein folds from their sequences. For some, AlphaFold is causing the merge of structural biology with bioinformatics. Here, using an in silico SMFS approach pioneered by our group, we investigate how reliable AlphaFold structure predictions are to investigate mechanical properties of Staphylococcus bacteria adhesins proteins. Our results show that AlphaFold produce extremally reliable protein folds, but in many cases is unable to predict high-resolution protein complexes accurately. Nonetheless, the results show that AlphaFold can revolutionize the investigation of these proteins, particularly by allowing high-throughput scanning of protein structures. Meanwhile, we show that the AlphaFold results need to be validated and should not be employed blindly, with the risk of obtaining an erroneous protein mechanism.

Mechanoactive proteins are essential for a myriad of physiological and pathological processes. Guided by the advances in single-molecule force spectroscopy (SMFS), we have reached a molecular-level understanding of how mechanoactive proteins sense and respond to mechanical forces. However, even SMFS has its limitations, including the lack of detailed structural information during force-loading experiments. That is where molecular dynamics (MD) methods shine, bringing atomistic details with femtosecond time-resolution. However, MD heavily relies on the availability of high-resolution structural data, which is not available for most proteins. For instance, the Protein Data Bank currently has 192K structures deposited, against 231M protein sequences available on Uniprot. But many are betting that this gap might become much smaller soon. Over the past year, the AI-based AlphaFold created a buzz on the structural biology field by being able to predict near-native protein folds from their sequences. For some, AlphaFold is causing the merge of structural biology with bioinformatics. Here, using an in silico SMFS approach pioneered by our group, we investigate how reliable AlphaFold structure predictions are to investigate mechanical properties of Staphylococcus bacteria adhesins proteins. Our results show that AlphaFold produce extremally reliable protein folds, but in many cases is unable to predict high-resolution protein complexes accurately. Nonetheless, the results show that AlphaFold can revolutionize the investigation of these proteins, particularly by allowing high-throughput scanning of protein structures. Meanwhile, we show that the AlphaFold results need to be validated and should not be employed blindly, with the risk of obtaining an erroneous protein mechanism.