DEL
DNA-encoded libraries are a high-throughput screening technology in which each compound is attached to a unique DNA sequence serving as a barcode. Billions of compounds are pooled and incubated with a target protein, after which sequencing reads out the enriched molecules, enabling compound screening in drug discovery.
Timeline (last 90 days)
Leash Bio uses a Lab-in-the-Loop closed-loop approach, combining DEL experiments with iterative AI model updates to run the cycle of experimental design and data analysis at computational speed.
DEL (DNA-encoded library) attaches a unique DNA barcode to each compound, performing billions of binding experiments simultaneously in a single tube, then reading out the enriched compounds by sequencing.
The DEL concept was founded by teams such as Affinity Pharmaceuticals in the 2000s and is now used to build billion-scale compound libraries.
Traditional screening methods such as ASMS and SPR typically stop at tens of thousands of compounds in throughput, whereas DEL can cover tens of billions.
DEL data can be used to detect whether an AI model achieves high scores by memorizing dataset biases (shortcut learning) rather than by learning chemical rules.
Leash Bio uses DEL data for model evaluation, which can identify models that score high on standard benchmarks but have poor generalization ability.
The unique value of DEL data lies in the fact that negative samples come from real experimental non-binding results and cover a broad chemical space; each round of selection amounts to an empirical test of the model's hypotheses, offering a test of true generalization closer to reality than synthetic benchmarks.
Leash Bio repositions DEL as AI training infrastructure, producing reasoning-trajectory data with experimentally validated labels, rather than as a traditional one-off screening tool.
DEL experiments involve long operational chains, high noise, and high failure rates, and issues such as library construction bias, nonspecific adsorption, and sequencing depth limitations can drown the real signal in noise.
All Facts (9)
Leash Bio uses a Lab-in-the-Loop closed-loop approach, combining DEL experiments with iterative AI model updates to run the cycle of experimental design and data analysis at computational speed.
50%UnverifiedDEL (DNA-encoded library) attaches a unique DNA barcode to each compound, performing billions of binding experiments simultaneously in a single tube, then reading out the enriched compounds by sequencing.
50%UnverifiedThe DEL concept was founded by teams such as Affinity Pharmaceuticals in the 2000s and is now used to build billion-scale compound libraries.
50%UnverifiedTraditional screening methods such as ASMS and SPR typically stop at tens of thousands of compounds in throughput, whereas DEL can cover tens of billions.
50%UnverifiedDEL data can be used to detect whether an AI model achieves high scores by memorizing dataset biases (shortcut learning) rather than by learning chemical rules.
50%UnverifiedLeash Bio uses DEL data for model evaluation, which can identify models that score high on standard benchmarks but have poor generalization ability.
50%UnverifiedThe unique value of DEL data lies in the fact that negative samples come from real experimental non-binding results and cover a broad chemical space; each round of selection amounts to an empirical test of the model's hypotheses, offering a test of true generalization closer to reality than synthetic benchmarks.
50%UnverifiedLeash Bio repositions DEL as AI training infrastructure, producing reasoning-trajectory data with experimentally validated labels, rather than as a traditional one-off screening tool.
50%UnverifiedDEL experiments involve long operational chains, high noise, and high failure rates, and issues such as library construction bias, nonspecific adsorption, and sequencing depth limitations can drown the real signal in noise.
50%