Abstract:Abstract We have developed an algorithm and implemented it in a software platform for the purpose of developing new anti-tumor drugs in the form of small molecules. In this study, we focused on generating molecules specifically for the treatment of lung cancer patients. To begin with, we employed deep learning (DL) techniques to evaluate the genes associated with poor clinical outcomes in lung cancer patients. By utilizing generative adversarial neural networks (GAN), we acquired additional patient data. The results of each experiment were presented as a list of genes ordered by their impact on the desired effect. We then intersected the lists of genes obtained from experiments with overall survival (OS) and progression-free interval (PFI) data. This allowed us to identify a set of genes whose expression was correlated with poor prognosis. In order to enhance the precision, we trained another DL model to distinguish between normal and tumor tissue based on gene expression. By doing so, we were able to identify the smaller set of genes that could be targeted. Subsequently, we developed a module that predicts the interactions between inhibitors and proteins. This involved representing protein amino acid sequences and chemical compound formulas in vector form, and a virtual screening of the Pubchem database. The implementation of the Drug-protein interactions module resulted in a dataset of 118,379 pairs, including 19,250 pairs describing compounds bound to proteins, and 99,129 precedents describing non-bound ones. DLwas applied, yielding a ROC-AUC of 0.86. Following the search for candidate molecules, we obtained 160,000 pairs with a predicted interaction probability above 0.99, as well as 2,921 pairs with probability of 1.0. Additionally, we created a DL-based module to predict the IC50 values in cell line experiments. Virtual pre-clinical trials were conducted using the selected inhibitors to identify relevant cell lines for subsequent laboratory experiments. Through this process, we obtained formulas for several molecules that demonstrated predicted binding to specific proteins. During the cell experiment emulation, our feature importance algorithm selected 129 genes. For the cell experiment emulation stage, we specifically chose interactions with a probability of at least 0.9. We prioritized molecules that acted on the minimum number of cell lines with a higher probability, thus ensuring higher specificity. Ultimately, we selected 5 small molecules as potential candidates, as well as certain cell lines for their validation. The NLP technologies utilized in this study demonstrated their effectiveness in processing tens of thousands of articles. The pipeline of methods presented in this paper lays the groundwork for automated AI-driven drug discovery. We have showcased the application of modern machine learning methods, particularly DL, as well as the methods used to prepare the initial data for the learning algorithms. The performance of these methods has been validated through cross-validation using data from publicly available sources. Citation Format: Dmitrii K Chebanov, Vsevolod A Misyurin, Nadezhda S Tatevosova. Deep learning-driven drug discovery: A breakthrough algorithm and its implication in lung cancer therapy development [abstract]. In: Proceedings of the AACR-NCI-EORTC Virtual International Conference on Molecular Targets and Cancer Therapeutics; 2023 Oct 11-15; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2023;22(12 Suppl):Abstract nr A014.

A versatile attention-based neural network for chemical perturbation analysis and its potential to aid surgical treatment: A experimental study

ComABAN: refining molecular representation with the graph attention mechanism to accelerate drug discovery

AttenSyn: An Attention-Based Deep Graph Neural Network for Anticancer Synergistic Drug Combination Prediction

A deep learning based multi-model approach for predicting drug-like chemical compound's toxicity

DeepDDS: deep graph neural network with attention mechanism to predict synergistic drug combinations

SAFER: sub-hypergraph attention-based neural network for predicting effective responses to dose combinations

Prediction of Adverse Drug Reactions by Combining Biomedical Tripartite Network and Graph Representation Model.

TCR: A Transformer Based Deep Network for Predicting Cancer Drugs Response

SynergyX: a multi-modality mutual attention network for interpretable drug synergy prediction

Predicting transcriptional responses to novel chemical perturbations using deep generative model for drug discovery

Deep fusion learning facilitates anatomical therapeutic chemical recognition in drug repurposing and discovery

DTSyn: a Dual-Transformer-based Neural Network to Predict Synergistic Drug Combinations

Equivariant Graph Attention Networks with Structural Motifs for Predicting Cell Line-Specific Synergistic Drug Combinations

Advanced graph and sequence neural networks for molecular property prediction and drug discovery

A Deep Neural Network for Predicting Synergistic Drug Combinations on Cancer

DeepCDR: a Hybrid Graph Convolutional Network for Predicting Cancer Drug Response

Abstract A014: Deep learning-driven drug discovery: A breakthrough algorithm and its implication in lung cancer therapy development

Network Representation Learning-Based Drug Mechanism Discovery and Anti-Inflammatory Response Against COVID-19

Accurate Clinical Toxicity Prediction using Multi-task Deep Neural Nets and Contrastive Molecular Explanations

Multi-layer graph attention neural networks for accurate drug-target interaction mapping

An Innovative Multi-Omics Model Integrating Latent Alignment and Attention Mechanism for Drug Response Prediction