Building Block-Based Binding Predictions for DNA-Encoded Libraries

Abstract

DNA-encoded libraries (DELs) provide the means to make and screen millions of diverse compounds against a target of interest in a single experiment. However, despite producing large volumes of binding data at a relatively low cost, the DEL selection process is susceptible to noise, necessitating computational follow-up to increase signal-to-noise ratios. In this work, we present a set of informatics tools to employ data from prior DEL screen(s) to gain information about which building blocks are most likely to be productive when designing new DELs for the same target. We demonstrate that similar building blocks have similar probabilities of forming compounds that bind. We then build a model from the inference that the combined behavior of individual building blocks is predictive of whether an overall compound binds. We illustrate our approach on a set of three-cycle OpenDEL libraries screened against soluble epoxide hydrolase (sEH) and report performance of more than an order of magnitude greater than random guessing on a holdout set, demonstrating that our model can serve as a baseline for comparison against other machine learning models on DEL data. Lastly, we provide a discussion on how we believe this informatics workflow could be applied to benefit researchers in their specific DEL campaigns.

Building Block-Centric Approach to DNA-Encoded Library Design

“…Further work is necessary to develop guidelines for three-cycle libraries or modes to identify covalent inhibitors, protein–protein interaction disruptors, or molecular degraders , to accommodate their expanded PCPs. Additional studies comparing different dimensionality reduction techniques for DEL analysis, such as TMAP, GTM, or UMAP, are also likely to be valuable. Finally, while model library size was constrained to 192 × 192 (∼37k numerical diversity), which is at least an order of magnitude smaller than some DELs, the field is migrating toward smaller libraries with more lead-like to drug-like properties. , These trends have accompanied experimental design innovations, enabling the detection of weaker binding events from lower cycle libraries through photoactivatable handles. − …”

Compositional Deep Probabilistic Models of DNA-Encoded Libraries

“…To that end, other methods have tackled this problem by utilizing the molecular structure via molecular fingerprints and graph neural networks. McCloskey et al 12 and Zhang et al 13 bin the count data and construct a classification problem based on their discretizations of the data. In particular, Zhang et al 13 propose exploiting the compositional structure of DELs for the extraction of enrichment signals, but unlike our work, no explicit generative models of this factorized representation or the improved likelihoods to deal with DEL data are built.…”

“…McCloskey et al and Zhang et al bin the count data and construct a classification problem based on their discretizations of the data. In particular, Zhang et al propose exploiting the compositional structure of DELs for the extraction of enrichment signals, but unlike our work, no explicit generative models of this factorized representation or the improved likelihoods to deal with DEL data are built. Other approaches formulate the problem as a latent-variable prediction task, maximizing the probability of observing the count data under some prescribed probability distribution such as the Poisson or negative binomial distribution. − Shmilovich et al extend the representation capabilities of models on DEL data by incorporating 3D docked poses to enhance the performance of models without requiring additional supervised validation data.…”