AI in Life Science: Weekly Insights
Weekly Insights | March 23, 2025
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:
Lab-in-the-loop: AI-Driven Antibody Design with Iterative Optimisation 💉🤖
Multimodal Mastery: AI Predicts Drug Combination Outcomes with Clinical Precision 💊🧬
TxAgent: Agentic system for drug discovery with 211 tools 💻🤖
Dive into these game-changing innovations and explore how they are transforming the biotech and healthcare landscapes!
Lab-in-the-loop: AI-Driven Antibody Design with Iterative Optimisation 💉🤖
Designing therapeutic antibodies is one of the most promising yet complex challenges in drug discovery; requiring optimization across binding affinity, expression yield, specificity, and developability. Traditional methods are laborious and slow, and many ML-driven solutions fall short of real-world requirements. Enter Lab-in-the-loop (LitL), a new AI-powered platform from Genentech that combines generative models, predictive oracles, and automated lab experiments in an iterative closed-loop system. By integrating machine learning with real-time experimental feedback, LitL produces fully viable, affinity-optimized antibodies - all without manual guidance.
🔨 Applications:
Therapeutic Antibody Discovery - Designs novel antibody variants with superior binding affinity, targeting clinically relevant antigens like EGFR, HER2, IL-6, and OSM.
Multi-Property Optimization - Simultaneously improves key developability traits like expression yield, stability, and specificity, ensuring candidates are ready for the clinic.
Structural & Functional Engineering - Combines ML and structural modeling to introduce mutations that enhance binding while preserving CDR loop integrity, verified by crystallography.
📌 Key Insights:
Generative Ensemble Approach - Uses multiple models (e.g. LaMBO-2, DyAb, PropEn) to generate antibody variants, enabling both exploration and exploitation of sequence space.
Active Learning Optimization - A property-ranking engine evaluates thousands of variants per round, selecting non-dominated designs that expand the Pareto frontier for binding and expression.
Real-Time Experimental Feedback - In each design round, new lab results retrain the models—improving performance with every cycle, and achieving up to 100× better binders.
Mechanistic Validation - Structural analysis shows that designed mutations introduce stabilizing interactions or modulate loop dynamics, providing interpretable gains in affinity.
Multimodal Mastery: AI Predicts Drug Combination Outcomes with Clinical Precision 💊🧬
Predicting clinical outcomes of drug combinations is critical for developing safe and effective therapies, yet existing models often fail to integrate diverse preclinical data types, limiting their accuracy. Introducing Madrigal, a multimodal AI model that unifyies structural, pathway, cell viability, and transcriptomic data to predict 953 clinical outcomes across 21,842 compounds—including novel and approved drugs. By handling missing data during training and inference, Madrigal outperforms single-modality models and enables early identification of adverse interactions and therapeutic synergies.
🔨 Applications:
Drug Repurposing - Identify novel combination of therapies to increase efficiency while reducing side effects of the treatment. It predicted safety profiles of PARP inhibitor combinations (e.g., olaparib + abiraterone vs. olaparib + paclitaxel) and identifies organ-specific toxicities (hematological, renal, hepatic)
Personalized Cancer Treatment - Integrates patient genomic profiles from acute myeloid leukemia (AML) samples and xenograft models to predict drug synergy and progression-free survival.
Transporter-Mediated Interactions - Flags risks for drugs sharing transporters (e.g., doxycycline-tacrolimus) and aligns with pharmacokinetic studies.
📌 Key Insights:
Unified data - Madrigal unifies multimodal inputs (structure, pathways, transcriptomics) via contrastive learning, creating aligned latent representations even with incomplete data.
Large data - Madrigal the effect efficacy and side effects of 238M possible pairs of compounds coming from 21,842 unique compounds
LLM Integration - Madrigal-LLM interprets free-text outcome descriptions (e.g., anemia risk) beyond predefined ontologies, enhancing flexibility for novel queries.
Real-World Validation - Correlates predictions with clinical trial data (e.g., neutropenia in PARPi combinations) and FDA safety databases (DrugBank, TWOSIDES), achieving 10.7% average improvement over baselines in adverse event prediction.
TxAgent: Agentic system for drug discovery with 211 tools 💻🤖
Precision therapeutics require multimodal adaptive models that generate personalised treatment recommendations. While a lot of data and tools exists, it is difficult and time consuming to run the tools and connect them. Introducing TxAgent, an AI agent that leverages multi-step reasoning and real-time biomedical knowledge retrieval across a toolbox of 211 tools to analyze drug interactions, contraindications, and patient-specific treatment strategies.
🔨 Applications:
Personalised Treatment Recommendations – TxAgent integrates real-time biomedical knowledge from FDA-approved drug labels, Open Targets, and other sources to tailor treatment strategies based on patient-specific factors, including age, genetic profile, and disease progression.
Drug-Drug Interaction Analysis – TxAgent assesses molecular, pharmacokinetic, and clinical interactions between drugs, identifying contraindications and potential adverse effects. For example, it flagged the interaction between Prozac (fluoxetine) and Xolremdi (mavorixafor) due to CYP2D6 inhibition
📌 Key Insights:
Multi-step reasoning optimised for drug discovery - TxAgent outperformed much larger LLMs in multi-step therapeutic reasoning, achieving up to 92.1% accuracy in drug selection, treatment personalisation, and therapeutic reasoning.
Tool-augmented AI – Interacts with 211 biomedical “tools”, with capabilities including finding information for all FDA-approved drugs, clinical insights and publications
Real-time retrieval – Queries latest treatment indications, regulatory approvals from continuously updated sources
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