Genome-wide investigation of gene-cancer associations for the prediction of novel therapeutic targets in oncology

Abstract

A major cause of failed drug discovery programs is suboptimal target selection, resulting in the development of drug candidates that are potent inhibitors, but ineffective at treating the disease. In the genomics era, the availability of large biomedical datasets with genome-wide readouts has the potential to transform target selection and validation. In this study we investigate how computational intelligence methods can be applied to predict novel therapeutic targets in oncology. We compared different machine learning classifiers applied to the task of drug target classification for nine different human cancer types. For each cancer type, a set of “known” target genes was obtained and equally-sized sets of “non-targets” were sampled multiple times from the human protein-coding genes. Models were trained on mutation, gene expression (TCGA), and gene essentiality (DepMap) data. In addition, we generated a numerical embedding of the interaction network of protein-coding genes using deep network representation learning and included the results in the modeling. We assessed feature importance using a random forests classifier and performed feature selection based on measuring permutation importance against a null distribution. Our best models achieved good generalization performance based on the AUROC metric. With the best model for each cancer type, we ran predictions on more than 15,000 protein-coding genes to identify potential novel targets. Our results indicate that this approach may be useful to inform early stages of the drug discovery pipeline.

PINNED: Identifying Characteristics of Druggable Human Proteins Using an Interpretable Neural Network

“…Li & Lai, 2007;Lin et al, 2019;Sikander et al, 2022;Sun et al, 2018;Zhu et al, 2009). Others only focus on a specific target or indication, such as oncology (Bazaga et al, 2020;de Falco et al, 2021;Dezső & Ceccarelli, 2020;Jeon et al, 2014), or ion channels (Huang et al, 2010). Restricting our focus to models which seek to assess the druggability of the entire proteome, we find that PINNED comfortably outperforms much of the prior literature in sensitivity, specificity, and AUC (Bull & Doig, 2015;Costa et al, 2010;Ferrero et al, 2017;Yao & Rzhetsky, 2008)…”

OncoRTT: Predicting novel oncology-related therapeutic targets using BERT embeddings and omics features

“…Finally, we averaged the AUC that is calculated for each fold for all ten folds. Furthermore, another evaluation setting used by (Bazaga et al, 2020), was also implemented to compare our results with this baseline method using their dataset and their procedure to perform a fair comparison, which is explained in more detail in the comparison section. Finally, we retrained our models using all positive and negative data samples in our datasets and then applied these models on new unseen test data to predict the labels of this new data (i.e., predict the novel therapeutic targets as positive data).…”

Omics Data and Data Representations for Deep Learning-Based Predictive Modeling

“…In [ 141 ], a DNN, along with the conventional ML models, utilized multi-omics data to predict the novel targets for a therapeutic treatment in the field of oncology. Starting from the lists of the genes that were either targeted by the FDA-approved drugs or those which, when they are mutated, may cause cancer, the researchers collected the data for the gene expression, the gene mutations (averaged over numerous patient samples for each cancer type), the gene essentiality (real-valued, mean sensitivity from knock-out experiments), and the gene interaction networks that were embedded via AE-based diffusion graphs [ 142 ].…”

“…In building a model to identify the venomous proteins from amino acid sequences [ 77 ], the researchers collected non-venomous samples by querying for the proteins that did not have the word “venom” on their metadata descriptions. In [ 141 ], the positive samples were data of genes that were targeted by drugs. For the negative samples, the researchers used a random subset, of the same size as the positive subset, of genes that were not listed as known therapeutic targets for FDA-approved drugs.…”