Gen-Drug
Sukrati Sharma, Kalyani Singh, Basu Dev Shivahare, Alavikunhu Panthakkan · 2026
Conventional drug discovery is time and money intensive as well as limited in its ability to scan the chemical space. The study presents GEN-DRUG (Generative Neural Drug Discovery and Optimization Platform), a generative artificial intelligence-based platform using powerful deep generative models to speed lead optimization, molecular screening, and de novo drug design around these restrictions. A Transformer-based activity predictor, an RL fine-tuner for pharmacokinetic and toxicity optimization, and a VAE-guided molecule generator form GEN-DRUG&s;s multi-stage workflow. More than 50 disease-relevant binding affinity labels are available with the system to train on more than 15 million compounds in the ChEMBL, ZINC, and PubChem databases.QED > 0.85 for more than 78% of the generated compounds, a validity rate of 99.2%, and 97.8% uniqueness score established that GEN-DRUG could generate novel, valid, and synthesizable drug-like molecules, as confirmed in silico. When benchmarked against 10 key protein targets such as EGFR, BRAF, and SARS-CoV-2 Mpro, GEN-DRUG produced high-affinity molecules with 34% more rapid convergence than comparisons GENTRL and MolGAN and improved docking score predictions by 22%. GEN-DRUG case study with kinase inhibitor showed that, as opposed to industry norm 6–12 months, GEN-DRUG discovered 7 new compounds with IC₅₀ < 50 nM, 3 of which achieved ADMET profiling and reached in vitro manufacture and testing in 42 days.The speeding up of the discovery timeline, widening chemical diversity, and targeting domain experts onto high-potential therapeutic ideas earlier, more effectively, and with better chances of success, GEN-DRUG shows the revolutionary potential of generative AI.