DNA double helix animation
Bridging AI and Chemistry to help saving lives one day!

Thibaud Southiratn

MSCA Doctoral Candidate (AI / Drug Discovery)

MolecularAI@AstraZeneca - LIAC@EPFL

Profile image

Work

PhD student based in Paris (currently in Sweden) with an interest in leveraging AI for health-related problems. I am driven by the potential of technology to make a positive impact on people's lives.

Right now, I am a MSCA Doctoral Candidate at AstraZeneca and EPFL working on Agentic AI for Drug Discovery.

Bio

2026 - NowConducting doctoral research at MolecularAI (AstraZeneca) and LIAC (EPFL), co-supervised by Samuel Genheden and Philippe Schwaller.
2021 - 2025Graduated from a dual-degree program with Seoul National University (BHI Lab) and Télécom SudParis in Computer Science.
2018 - 2021Prepared for the French "Grandes Écoles" entrance exams. Got strong foundations in Mathematics, Physics, and Chemistry.
2018Graduated from high-school with highest honors (Scientific track).
2002Born in Bourg-la-Reine, France.

Publications

CombiMOTS: Combinatorial Multi-Objective Tree Search for Dual-Target Molecule Generation

ICML'25

Thibaud Southiratn, Bonil Koo, Yijingxiu Lu, Sun Kim

Deep Learning vs Classical Methods in Potency & ADME Prediction: Insights from a Computational Blind Challenge

ACS - JCIM

Yaelle Fischer, Thibaud Southiratn, Dhoha Triki, Ruel Cedeno

Web-based Exploratory Data Mining System for Analyzing the Gene-level Relationship between Intratumoral Heterogeneity of Promoter DNA Methylation and Drug Response

2024 KCC

Tae Hoon Kweon, Bonil Koo, Sungjoon Park, Thibaud Southiratn, Sun Kim

News & Awards

📰 [12/2025] The article is live on JCIM !

📰 [10/2025] New preprint on Potency & ADME prediction is out!

🏆 [10/2025] I ranked 1st in Owkin's Decoding Biology Hackathon!

📰 [08/2025] Graduated from SNU :( See you 서울!

📰 [05/2025] Yaaay!! My first paper got accepted to ICML'25! > Vancouver gogo 🇨🇦!

Projects

Selective CDK7 Inhibitor Generation

PARETO OPTIMIZATION • MONTE-CARLO TREE SEARCH • PROPERTY PREDICTION

• Adapted CombiMOTS to attempt unveiling molecules biochemically active to CDK7 & inactive to CDK1-2-5-9-12-13.

• Identified potent candidates with motifs/warheads (acrylamide, chloroacetamide) found in relevant literature.

Efficient Molecule Captioning

TRANSFORMERS • CHEMICAL LANGUAGE MODELS

• Fine-tuned a Chemical LM (Text+ChemT5) to improve performance (up to +5.4%) on the "mol2text" task.

• Adapted an implementation of Speculative Decoding to infer captions faster (+36.5%) without changing output distribution.

How I spend my time

Photography for creativity (yes it's cliché), gaming to remain competitive and working out to stay in shape!

On the web

© 2026 Thibaud Southiratn. All Rights Reserved.
Based on design by Takuya Matsuyama