#8: Life Science x AI
Welcome back to your weekly dose of AI news for Life Science!
Here’s what we have for you this week:
Streamlining Drug Discovery with SPARROW 🧪
The impact on clinical success from the 23andMe cohort 💊
Life Science tools of the week 🛠️
We have started working on an AI Scientist to design, execute and troubleshoot scientific experiments …. join the waiting list to be the first one to try it!
Streamlining Drug Discovery with SPARROW 🧪
You struggle to decide which molecule you want to move forward in your pipeline? Sure, generative models can help iterating and proposing molecules, but often these molecules are impractical or very expensive to synthesize! SPARROW comes to the rescue!
📌 SPARROW algorithm balances the growth of predicted properties with forward synthesis and backward analysis, maintaining the cost of synthesized chemicals in line with its expected information value.
📌 In a nutshell, SPARROW select the strategy path to accomplish more cumulative rewards while avoiding more synthetic costs and risking failure of the reaction to the strategy.
📌 SPARROW also incorporates the cost savings associated with common intermediates and starting materials in batch synthesis.
Moreover … SPARROW is completely open-source!
The impact on clinical success from the 23andMe cohort 💊
It’s always interesting to read on how genetics is helping finding the right targets and bringing drugs to the market faster, especially when 90% of drug candidates fail in clinical trials. Today we read a nice article from 23AndMe about the impact of their cohort on clinical trials success
📌 23andMe has approximately 15 million individuals with genotype and phenotype data (30x larger than the UK Biobank), of which ~80% consent to participation in research.
📌 Genetic data from 23andMe has increased the number of target-indication pairs in clinical trials by 60% compared to all publicly available GWAS datasets.
📌 Drug programs supported by rare and large effect genetic associations have a 3-4 times greater likelihood of approval compared to those with common variant associations with small effects.
Life Science tools of the week 🛠️
1/ HELM-GPT - Macrocyclic peptide design
Today, drugs can target ~20% of all disease-relevant human proteins. Why? Cause a lot of disease-causing proteins are intracellular and “undruggable” as they do not contain well-defined binding pockets for small molecules and are inaccessible to biologics (eg. antibodies) due to the obstacle of the cell membrane. Many promising therapeutic modalities are there, including macrocyclic peptides which can target intracellular proteins and can bind flat protein surfaces.
Introducing HELM-GPT, an open-source tool that applies GenAI to design macrocyclic peptides!
📌 HELM-GPT was trained on ChEMBL (22,040 cyclic peptides) and CycPeptMPDB (7451 cyclic peptides) databases, along with KRAS-related peptides from patents.
📌 HELM-GPT beats several models in optimising macrocyclic peptides for cellular permeability
📌 Some limitations still exists, including co-optimization of unrelated properties, such as cell permeability and KRAS binding affinity
2/ InfoAlign - Learning Molecular Representation in a Cell
Introducing Information Alignment (InfoAlign), the latest approach for learning molecular representations! Here the quick insights
📌 InfoAlign uses a “information bottleneck” graph approach which discards redundant information while retaining sufficient data to align with different biological features
📌 InfoAlign supports multiple features, including activity classification and ADMET data
📌 InfoAlign outperforms up to 19 baseline methods across four datasets, showing a +10.58% and +6.33% improvement in classification and regression tasks
3/ Nuclei - Clinical Imaging
Researchers from Stanford published on Nature Biomedical Engineering a new tool to help clinician leverage AI for image segmentation!
Introducing nuclei.io!
📌 Nuclei.io allows users to without coding skills to easily deploy a local platform for cell image segmentation with AI features, such shape and morphology extractions
📌 Users can easily select a group of cells and assign to the eg. tumour category, and the tool in real-time will re-color all similar cells in the slide! What’s more, users can in real-time change label more cells as tumour ones and the AI algorithm will re-train and re-color the whole slide
📌 The result so far? It saved clinicians 20% of time for finding lymph node metastasis!
BITE-SIZED COOKIES FOR THE WEEK 🍪
Are you lost with all the Pharma deals and M&A in the past years and would like to understand where to focus? Fierce Biotech have you covered with the article on the smartest pharma acquisitions!
New database with 451,065 Major Histocompatibility Complex peptide epitopes, each with experimental evidence for MHC binding, along with detailed information on human leukocyte antigen allele specificity, source peptides, and references to original studies!
Don’t forget to take our short survey to help us understand how you currently use AI in your day-to-day!