<p>    </p> <p>DNA-encoded chemical libraries (DELs) are collections of DNA-linked small molecules, where each synthetic small molecule is covalently attached to a unique DNA barcode that encodes its identity. This technology harnesses the power of organic chemistry and genetics, which extends the application of molecular evolution and natural selection to the discovery of specific small molecules binders to protein targets of interest. Rather than discretely screening individual molecules, up to billions of DNA-encoded small molecules can be assessed collectively by a selection assay in a single tube. As a result, the high sensitivity, low cost, and unprecedented level of molecular complexity of DELs allow rapid generation of novel bioactive compounds. While powerful, this approach has its own limitations, including limited target scope and selection strategies. Currently, DEL targets have been largely limited to biochemically purified proteins and used in affinity-based selections assays. In the first area of this work, we address both these limitations by capitalizing on the power of affinity labeling. This allows DELs to be applied to protein targets within and on living cells and expands the power of DNA-encoding to the identification of small molecules with specific biological functions beyond binding. </p> <p>In the second area, we harnessed affinity labeling and DNA sequence analysis to develop multiplexed small molecule ligand binding assays. This method is the initial demonstration of split-and-pool ligand binding assays using DNA-linked small molecule probes. We used this approach in a high-throughput screening campaign to identify selective inhibitors by screening 1000 compounds against 5 bromodomain proteins concurrently. In addition, this approach was utilized to rank order the affinity of a 96-member library of DNA-linked ligands to a protein simultaneously, which significantly increases the throughput of ligand binding assays while keeps the cost low. </p> <p>Lastly, we developed proximity-induced selection assays to enrich ligands from DELs. This approach involves uncaging or installation of a biotin purification tag on the DNA construct either through photo-deprotection of a protected biotin group using a light emitting protein tag or by amine acylation using an engineered biotin ligase. Compared to affinity labeling-based selection approaches, this approach results in improved recovery of ligands and, at the same time, removes the onerous requirement of protein purification. The enzyme-mediated proximity labeling approach should serve as a convenient tool for molecular discovery with DELs. </p>

The split-and-pool method has been widely used to synthesize chemical libraries of a large size for early drug discovery, albeit without the possibility of meaningful quality control. In contrast, a self-assembled DNA-encoded chemical library (DEL) allows us to construct an m x n-member library by mixing an m-member and an n-member pre-purified sub-library. Herein, we report a trio-pharmacophore DEL (T-DEL) of m x l x n members through assembling three pre-purified and validated sub-libraries. The middle sub-library is synthesized using DNA-templated synthesis with different reaction mechanisms and designed as a linkage connecting the fragments displayed on the flanking two sub-libraries. Despite assembling three fragments, the resulting compounds do not exceed the up-to-date standard of molecular weight regarding drug-likeness. We demonstrate the utility of T-DEL in linker optimization for known binding fragments against trypsin and carbonic anhydrase II and by de novo selections against matrix metalloprotease-2 and −9.

DNA-encoded chemical libraries (DEL) allows an exhaustive chemical space sampling with a large-scale data consisting of compounds produced through combinatorial synthesis. This novel technology was utilized in the early drug discovery stages for robust hit identification and lead optimization. In this project, the aim was to build a Machine Learning- based QSAR model with conformal prediction for hit identification on two different target proteins, the DEL was assayed on. An initial investigation was conducted on a pilot project with 1000 compounds and the analyses and the conclusions drawn from this part were later applied to a larger dataset with 1.2 million compounds. With this classification model, the prediction of the compound activity in the DEL as well as in an external dataset was aimed to be analyzed with identification of the top hits to evaluate model’s performance and applicability. Support Vector Machine (SVM) and Random Forest (RF) models were built on both the pilot and the main datasets with different descriptor sets of Signature Fingerprints, RDKIT and CDK. In addition, an Autoencoder was used to supply data-driven descriptors on the pilot data as well. The Libsvm and the Liblinear implementations were explored and compared based on the models’ performances. The comparisons were made by considering the key concepts of conformal prediction such as the trade-off between validity and efficiency, observed fuzziness and the calibration against a range of significance levels. The top hits were determined by two sorting methods, credibility and p-value differences between the binary classes. The assignment of correct single-labels to the true actives over a wide range of significance levels regardless of the similarity of the test compounds to the training set was confirmed for the models. Furthermore, an accumulation of these true actives in the models’ top hit selections was observed according to the latter sorting method and additional investigations on the similarity and the building block enrichments in the top 50 and 100 compounds were conducted. The Tanimoto similarity demonstrated the model’s predictive power in selecting structurally dissimilar compounds while the building block enrichment analysis showed the selectivity of the binding pocket where the target protein B was determined to be more selective. All of these comparison methods enabled an extensive study on the model evaluation and performance. In conclusion, the Liblinear model with the Signature Fingerprints was concluded to give the best model performance for both the pilot and the main datasets with the considerations of the model performances and the computational power requirements. However, an external set prediction was not successful due to the low structural diversity in the DEL which the model was trained on.