Combining DEL and AI for hit finding: hype or reality?

Posted on Nov 13, 2025

This was the subject of the recent DREAM Target 2035 Drug Discovery Challenge organized by Structural Genomics Consortium (SGC), Pfizer, IBM Research, and its partners.

The aim was to build Artificial Intelligence (AI)/Machine Learning (ML) models trained on DNA-Encoded Libraries (DEL) data and then predict confirmed hits from an unseen unrelated library (ASMS).

Ruel Cedeno, from our Cheminformatics team, joined over 231 scientists from both industry and academia worldwide (GSK, ChemSpace, New York University, Harvard University, etc).

Novalix’s DEL-ML approach amongst the top performers

We are happy to share that our DEL-ML approach at Novalix was among the top performers (fingerprint round):

# 1 in ROC-AUC metric (how well the model ranks binders above non-binders)
# 2 in number of distinct chemotypes correctly predicted
# 3 in number of binders correctly predicted
# 5 in overall evaluation metric

Link to the leaderboard: https://www.synapse.org/Synapse:syn65660836/wiki/632249  

This outcome demonstrates the powerful combination of DEL and ML, allowing hit finding and expansion beyond the DEL chemical space.

To explore how we can work together to achieve new hits using our globally competitive DEL-ML pipeline, let’s talk.

For more details about our drug discovery services, click here.

Looking for more information on drug discovery?