Abstract:
Cancer remains a major global health challenge, necessitating advanced computational resources to support effective treatment and disease management. These resources can assist researchers in the early detection of cancer and the screening of therapeutic molecules. This study presents a suite of machine learning-based in-silico models and resources developed to facilitate cancer diagnosis and therapy. First, we developed a highly accurate classifier based on a ten-gene panel selected using propensity indices. This model reliably distinguishes prostate cancer from normal samples, offering a potential non-invasive alternative to conventional diagnostic procedures. In recent years, there has been a notable shift in drug discovery from small molecules to peptide-based therapies. To support this transition, we developed THPdb2, an updated repository of FDA-approved therapeutic peptides. Building on this, we introduced THPPred, a machine learning-based tool that predicts novel druggable peptides by analysing their physicochemical, molecular, and sequence features, thereby streamlining therapeutic screening. To further support peptide-based immunotherapy, we developed IL13Pred, a predictive model that identifies IL-13-inducing peptides, aiding in the design of cytokine-based cancer immunotherapies. Additionally, for small-molecule drug discovery, we created NfκBin, an in-silico model that predicts NF-κB inhibitors using molecular fingerprints, accelerating the identification of compounds targeting this key oncogenic pathway. Collectively, these integrated models and databases form a comprehensive computational pipeline for cancer biomarker discovery and therapeutic screening. By combining gene expression analysis, peptide-based drug prediction, and small-molecule screening, this framework advances personalized diagnostics and treatment strategies in oncology. Mostly resources developed in this study are publicly accessible via a web-based platform and GitHub.
