Spurti U Akki , Kirill Novikov , Divya Kanichar , Patrick J McEnaney , Tracy Chan , Vincent Lee , Jacob H Pachter , Joshua L Turnbull , Nikhil Jain , Virja A Pandya , Nathaniel H Stanley , Dong Min Mun , Christoph Klein , Boris P Belotserkovskii , R. Edward Watts
ChemRxiv
DOI: 10.26434/chemrxiv.15007812/v3
Abstract
DNA-Encoded Libraries (DELs) are indispensable high-throughput platforms in drug discovery, but the widely adopted split-and-pool synthesis method is fundamentally limited: the DNA merely records, rather than directs, the steps of chemistry. This lack of programmatic control imposes mandatory downstream chemistry congruency, restricts structural diversity, and inevitably generates a combinatorial explosion of un-drug-like molecules, hindering both hit-finding and the acquisition of structurally informative data required for machine learning (ML) models. To resolve these core constraints, we developed Programmed DEL Synthesis (PDS), a robust, industrially applicable platform for non-combinatorial DEL construction in which the DNA actively directs the synthesis. PDS is built upon a sort-and-pool mechanism expanded by three key innovations: sequence-specific installation of first building blocks using charged carrier codons, a multinomial encoding strategy that enables libraries exceeding 20 billion unique encodings in three chemical steps, and non-combinatorial germline assembly that allows for the programmatic omission of undesirable building block combinations. We demonstrate the platform with Library i01, a three-step, threepathway library employing 137 trifunctional scaffolds. The design successfully programmed the systematic merging and un-merging of three mutually incompatible reaction pathways: Acid→Aldehyde→Amine, Amine→Aldehyde→Acid, and Aldehyde→Acid→Amine within a single synthesis. Pilot affinity selections against Carbonic Anhydrase (BCA-II) validated the high fidelity of DNA encoding and efficient building block installation across all steps and pathways, confirming the platform's robust utility for selection. PDS transforms chemical congruency and drug-likeness into programmable properties, offering a precise, efficient solution that improves library quality without sacrificing chemical space. Crucially, PDS enables the construction of structurally tailored 2nd Generation DELs needed to provide the high-resolution data required for productive ML-driven active learning cycles, establishing a critical operational link for modern drug development.