In this issue:
Welcome back to your weekly dose of AI news for Life Science!
This week, we have some exciting new models lined up for you:
KGWAS: AI-powered Knowledge Graph to enhance genomic studies for uncommon diseases 💿
GenMol: Fragment-constrained generative chemistry for small molecule design 💻
Dive into these game-changing innovations and explore how they are transforming the biotech and healthcare landscapes!
KGWAS: AI-powered Knowledge Graph to enhance genomic studies for uncommon diseases 💿
Genome-wide association studies (GWASs) have identified tens of thousands of disease associated variants and provided critical insights into developing effective treatments. However, limited sample sizes have hindered the discovery of variants for uncommon and rare diseases. Introducing KGWAS, a novel deep learning method that leverages a massive functional knowledge graph across variants and genes to improve detection power in small-cohort GWASs significantly!
📌 Key Insights:
KGWAS is the largest AI model that integrates >10 millions of multi-modal and multi-scale functional genomics data to improve GWAS power by 100% while discovering novel disease-critical variants, genes, cells, and networks!
KGWAS identified up to 100% more statistically significant associations than state-of-the-art GWAS methods and achieved the same statistical power with up to 2.67× fewer samples.
KGWAS uncovered up to 47% more links to 554 uncommon diseases and an impressive 80% more associations with 142 rare diseases in the UK Biobank
KGWAS Knowledge Graph is built on top of a large dataset of ~8.6M variant-to-gene, ~2.3M gene-to-gene and ~115k gene-to-program associations
Bio-xLSTM: Generative modelling, representation and in-context learning of biological and chemical sequences 🧬
When Transformer-based models are increasingly used to treat and predict biological sequences, most of the models suffer from the sequence's length, which complicates their use in strong biological cases. A different type of AI architecture (xLSTMs) has been shown to perform favourably compared to Transformers to handle long-range dependencies … which is essential for biology and chemistry! Introducing Bio-xLSTM, a family of models (DNA-xLSTM, Chem-xLSTM and Prot-xLSTM) applied to accurately predict genomics, chemical and protein sequences!
📌 Key Insights:
All the models performed quite well across a variety of benchmark!
Can perform in-context learning for proteins and small molecules, meaning it can easily learn and model new relationship in the data
While promising, this work is still early on and the models and generalisability are constrained by the training dataset
GenMol: Fragment-constrained generative chemistry for small molecule design 💻
Drug discovery is a complex process that involves multiple scenarios and stages, such as fragment constrained molecule generation, hit generation and lead optimisation. However, existing molecular generative models can only tackle one or two of these scenarios and lack the flexibility to address various aspects of the drug discovery pipeline. Introducing Generalist Molecular generative model (GenMol) from NVIDIA, a versatile framework that addresses these limitations!
📌 Key Insights:
GenMol is able to effectively and efficiently generate molecules
under settings that simulate a variety of drug discovery problems, such as
De Novo Generation,
Fragment-constrained Generation,
Goal-directed Hit Generation
Goal-directed Lead Optimisation
GenMol outperform previous models across most scenarios, including outperforming REINVENT for goal-directed hit generation
GenMol was trained with minimal cost - only 8 NVIDIA A100 GPUs for 6 hours!
Did you find this newsletter insightful? Share it with a colleague!
Subscribe Now to stay at the forefront of AI in Life Science.
Connect With Us
Have questions or suggestions? We'd love to hear from you!
📧 Email Us | 📲 Follow on LinkedIn | 🌐 Visit Our Website