AI 驱动的药物发现：借助 NVIDIA 和 Databricks 加速分子洞察.pdf

1、AI-Driven Drug Discovery:Accelerating Molecular Insights With NVIDIA and DatabricksKaruna NadadurSrijit Chandrashekhar NairIntroduction and ContextWhat is BioNeMo?Why Databricks is THE platform?DEMOHow BioNeMo works on Databricks Call To ActionQ&AAgendaAgendaWhat if,as a scientist-I need to fold pro

2、tein sequences or analyze molecular sequence quickly.-Quickly finetune a Pretrained protein fold and molecular analysis models with my proprietary data.-My data is in Databricks Lakehouse and I dont want to move my data around.-My work is highly protected Intellectual Property,all assets need to be

3、strictly governed.-I want my experiments to be tracked easily so that I can provide evidence of my research.-I dont want to spend a lot of time configuring GPUs and clusters.NVIDIA BioNeMo framework on DatabricksThe next 30 years of scientific innovation is programmable biologyThe quest to generate

4、biology and chemistry that meets our needsProtein structuresChemical structuresMALLHSARVASGVASAFHPGLAKGASARASSWWAHVEAMPPDPILGVTEASKRDTNSKKMNLGVGProtein sequencesCC(=O)OC1=CC=CC=C1C(=O)OChemical sequencesMALLHSARVASGVASAFHPGLAKGA+CC(=O)OC1=CC=CC=C1C(=O)OBound sequencesBound structuresNVIDIA BioNeMoBu

5、ild,Optimize and Deploy Foundation Models for Computer-Aided Drug DiscoveryBuild/CustomizeBiomolecular GenerationProtein Structure PredictionBiomolecular Property PredictionProtein Binder DesignProtein-Ligand DockingDeployOptimized ModelsDomain-Specific ToolingScalable TrainingNVIDIA DGX CloudBioNeM

6、o FrameworkBioNeMo NIMsOriginally-published models vs.BioNeMo Framework implementationsOriginal models from publicationsBioNeMo implementationPurposeDesigned primarily for proof-of-concept workloads such as inference and fine-tuning on small datasets Optimized extensively for enterprise production w