Related Experiment Video
Updated: Jul 2, 2026
07:28
Identification of Functionally-Relevant Lentivirus Integration Sites in an Insertional Mutagenesis Cell Library
Published on: January 10, 2025
Evaluation of DNA encoded library and machine learning model combinations for hit discovery
Sumaiya Iqbal1,2,3, Wei Jiang4, Eric Hansen4
1Broad Institute of MIT and Harvard, Center for the Development of Therapeutics, Cambridge, MA, 02142, USA. sumaiya@broadinstitute.org.
Npj Drug Discovery
|June 30, 2026
Summary
DNA-Encoded Library (DEL) technology combined with Machine Learning (ML) accelerates drug discovery. This study validated DEL+ML for identifying therapeutic targets, confirming predicted binders with high accuracy.
Area of Science:
- Medicinal Chemistry
- Computational Chemistry
- Drug Discovery
Background:
- DNA-Encoded Library (DEL) technology enables rapid screening of vast compound libraries.
- Machine Learning (ML) models can be developed using DEL data for virtual screening.
- Traditional drug discovery methods are often time-consuming and expensive.
Purpose of the Study:
- To comparatively assess the DEL+ML pipeline for hit discovery.
- To evaluate the performance of fifteen DEL+ML combinations across three DELs and five ML models.
- To identify orthosteric binders for therapeutic targets Casein kinase 1α/δ (CK1α/δ).
Main Methods:
- Utilized three distinct DEL libraries and five different ML models.
- Applied ML models to predict binders and non-binders for CK1α/δ.
- Validated predictions using biophysical assays.
Main Results:
- Achieved 10% confirmation rate for predicted binders and 94% for predicted non-binders.
- Identified two nanomolar binders (187 nM and 69.6 nM) for CK1α/δ.
- Demonstrated the effectiveness of the DEL+ML approach in hit discovery.
Conclusions:
- The DEL+ML paradigm is a powerful strategy for efficient hit discovery.
- Chemical diversity in training data is crucial for ML model performance.
- ML model generalizability is more important than accuracy for DEL applications.
