ML methods to accelerate the drug discovery cycle

on YouTube ↗31m

" summary

Built a Janssen knowledge graph on AWS that links molecules, assays, lab notebooks and regulatory docs using Amazon Neptune so teams query unified records instead of silos; an early POC proves discoverability and end‑to‑end asset tracking. A novel ML model LM‑GVP combines a protein language model with a 3D graph neural network, trains on ~2.1B sequences (293B amino acids), achieves state‑of‑the‑art function prediction with residue‑level interpretability. Quantum optimization experiments using D‑Wave/qbsolve show faster solve times and better fitness on 4‑mer and 14‑mer peptide design.

AI-written summary · about

" chapters (21)

1. 0:00Intro
2. 0:48Machine learning in healthcare and life science
3. 3:50Discovery and development: Candidate generation
4. 4:45Preclinical research: In vivo/vitro studies
5. 5:39Clinical research: Clinical trials Feasibility of clinical trials
6. 6:29Drug discovery: Candidate screening
7. 7:38Janssen knowledge graph: Business case
8. 8:46Knowledge graph usage patterns
9. 10:01Knowledge graph connectivity 2
10. 11:04Janssen knowledge graph architecture
11. 12:24LM-GVP: sequence and structure informed molecule attribute prediction
12. 13:55LM-GVP provides residue-level interpretability
13. 17:14Key business use cases enabled by the AWS knowledge graph platform
14. 18:19Building & leveraging institutional knowledge
15. 19:39NLP-enabled electronic laboratory notebook data mining
16. 20:20High throughput antibody discovery
17. 24:22Resolving protein structure
18. 25:19The protein design problem
19. 27:44Computing infrastructure
20. 28:24Solver performance
21. 30:02Summary

" related talks

• 19mQuantum-Centric Data Driven R&D with focus on Life ScienceIftikhar Ahmed · Bitkom Quantum Summit 2023
• 46mMachine Learning-based Design of Proteins and Small MoleculesJennifer Listgarten · Emerging Challenges in Deep Learning (Simons Institute workshop, 2019)
• 04mCustomer SpotlightSairam Menon · AWS Silicon Innovation Day 2023
• 30mMachine Learning: A New Approach to Drug DiscoveryDaphne Koller · Women in Data Science (WiDS) Stanford 2020
• 23mUsing Graph Nets (GNs) to predict molecular propertiesChaya D Stern, Yuanqing Wang · PyData New York City 2019
• 30mLeveraging Molecular Data for Drug DiscoveryHaiyan Huang · Computational Challenges in Very Large-Scale 'Omics'
• 11mImproving Drug Discovery with Machine Learning...Simon Axelrod · PyTorch Conference 2025
• 41mAccelerated Discovery to Knowledge using Machine LearningDr. Abhishek Singh · Thoughtworks E4R Symposium 2020