Interpretable Machine Learning for Molecular Property Prediction: Bridging Chemistry and AI

[M4Marvin]

Title

Talk Proposal - Interpretable Machine Learning for Chemistry

Describe your Talk

Machine learning (ML) has emerged as a powerful tool for predicting molecular properties, enabling breakthroughs in drug discovery, materials science, and computational chemistry. However, the lack of interpretability in many ML models poses a significant challenge, especially in chemistry research where understanding the underlying mechanisms is critical for validation and regulatory compliance. This workshop will address this challenge by combining interpretable machine learning with physics-based approaches to create accurate, efficient, and transparent models for molecular property prediction.

Participants will learn how to build and interpret ML models using minimal descriptors, enabling efficient predictions while maintaining interpretability. The workshop will cover state-of-the-art techniques such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to explain model predictions, as well as the integration of physics-based models to enhance accuracy. Hands-on sessions will guide participants through the development of ML models for predicting key molecular properties, such as hydration free energies and binding affinities, using tools like XGBoost, Random Forest, and PyTorch.

By the end of the workshop, participants will have gained practical experience in:

1. Building and validating ML models for molecular property prediction.
2. Applying interpretability techniques to understand and explain model predictions.
3. Integrating physics-based approaches with machine learning for enhanced accuracy and transparency.

This workshop is designed for chemists, computational researchers, and drug discovery professionals who want to leverage the power of interpretable machine learning in their work. No prior ML experience is required, as the workshop will provide a comprehensive introduction to the tools and techniques needed to get started.

Pre-requisites & reading material

Basic Python experience.

Time required for the talk

45

Link to slides/demos

Slides

About you

I am a computational scientist and full-stack architect with a dual degree in Computer Science and Computational Biology from Jawaharlal Nehru University. My research focuses on developing innovative machine learning solutions for complex biological systems, including protein-ligand interactions and molecular property prediction. I have published in prestigious journals such as The Journal of Physical Chemistry B and developed state-of-the-art tools like HAC-Net for protein-ligand binding affinity prediction. I am also the founder of a web development consultancy, where I build secure, high-performance platforms for academic and enterprise clients. I am passionate about bridging the gap between computational research and accessible technological implementations.

Availability

Late February to march 2025

Any comments

My times are flexible as i have my own firm. I can make time if i am informed a at least a week prior

[Animesh-Ghosh]

Hi @M4Marvin, could you update your initial comment and share a link to the slides?

It's fine even if it's a very basic slide which outlines the structure of the talk.

[M4Marvin]

Sure I can make some basic slides and send them by today.

[M4Marvin]

I have updated the initial comment

[Daksh777]

Hey @M4Marvin
The slides look pretty good to me and the talk is very unique too :)

Would you be available on 16th for our meetup? It's going to be in IIIT Delhi

[Schefflera-Arboricola]

isn't it on 15th?

[M4Marvin]

I am free on the 16th this month. Would love to be a part of the event.

[Daksh777]

@Schefflera-Arboricola we had to shift it to 16th because there were multiple meetups happening on 15th and IIIT-D was happy to host us on 16th :)

@M4Marvin awesome! could you email me your phone/WhatsApp number so that I can coordinate with you for the venue and timings for the day of the event? My email is [email protected]