In Patil et al. two network-based approaches are demonstrated to identify driver gene patterns among tumor samples from the Pan-Cancer Analysis of Whole Genomes. The first approach used a sequence similarity network (SSN) called the Directed Weighted All Nearest Neighbors (DiWANN) network, a more computationally efficient model than more traditional all-to-all distance matrices, that links each node only to the sequence that is closest to it by edit distance. The time complexity of generating the SSN was reduced by employing DiWANN, and additional refinement and data reduction allowed this analysis to be performed using only about 2,200 nucleotides over the conventional 300,000 nucleotides, yet still produced robust results that identified both exclusive and co-occurring driver genes in specific cancer types. These results were supported by their second approach, a bipartite network analysis that showed specific pairs of genes important in those specific cancer types.

Karampuri and Perugu delve into Machine Learning (ML) and Deep Learning (DL) approaches to developing structure-activity relationships between quantitative representations of a molecule’s structure and their biological activity. This Quantitative Structure-Activity Relationship (QSAR) model employs molecular descriptors, such as geometric, topological or physiochemical characteristics, and can be used to predict the biological activity of novel structures, or the potential efficacy of novel combinations of structures in combinatorial chemotherapy. In this work, the authors utilize the GDSC 2 (Genomics of Drug Sensitivity in Cancer drug Combinations) database to establish combinatorial QSAR models with 11 common regression-based ML and DL algorithms, demonstrating a particularly high prediction performance in several models as measured by R 2 and RMSE when comparing the QSAR model predictions to actual combinatorial IC50 values for cancer treatments.

While multiple sequence alignments (MSAs) have long been a staple of genomic analysis, they are often limited to visualizing alignment blocks that comprise only fragments of individual genes. Therefore it can be difficult to examine features like duplications or translocation of a sequence with traditional visual approaches like parallel coordinate views or dot plots. Schebera et al. examine genome-wide multiple sequence alignments (gMSAs), and propose a methodology by which to examine longer sequences without breaking them into shorter sequence intervals, thereby preventing the loss of order context, and allowing further investigation into pangenome to pangenome comparisons. Their graph layout framework algorithm is based on the Sugiyama framework, but where that framework was applicable to directed acyclic graphs (DAGs), the gMSA graph is a multi-graph with routing for multiple edges that reduces edge intersections and promotes high readability.

Bayesian networks (BNs) represent probabilistic relationships through DAGs, enabling visualization of complex systems and identifying causality between variables. In their study, Waddell et al. use BN modeling to show the link between MRI-derived deposition of body fat and the likelihood of hospitalization in COVID-19 patients. Besides this significant increase in hospitalization risk, the BN modeling also demonstrated the greatest risk to patients with higher amounts of visceral adipose tissue and liver fat. These results provide a model for potential examinations into other associations between patient factors and potential hospitalization risks using BN modeling.