We develop new genAI methods, including diffusion, reinforcement learning and autoencoders, to generate new drug and vaccine candidates.

A Deep Generative Model for Molecule Optimization via One Fragment Modification

a. The Modof-encoder. Modof first generates atom embeddings of Mx/My over molecular graphs Gx/Gy using GMPNs, as well as node embeddings over corresponding junction trees Tx/Ty using TMPNs. The difference between Tx and Ty at the disconnection site (circles in Tx/Ty) is encoded (DE) into hxy and h+xy, which then construct two normal distributions zxy and z+xy. b. The Modof-decoder. Using zxy, Modof conducts disconnection site prediction (DSP) to identify site nd. At neighbours of nd, Modof conducts removal fragment prediction (RFP) to remove the fragment at nd. Then, Modof produces an intermediate representation (IMR) of the remaining scaffold (G*,T*). Over (G*, T*), Modof performs new fragment attachment (NFA) by interactively performing child node connection prediction (NFA-cp), child node type prediction (NFA-ntp) and attachment point prediction (NFA-app) to optimize Mx. In molecule representations, substructures in molecular graphs and their corresponding nodes in junction trees are coded in the same colours.

AbstractMolecule optimization is a critical step in drug development to improve the desired properties of drug candidates through chemical modification. We have developed a novel deep generative model, Modof, over molecular graphs for molecule optimization. Modof modifies a given molecule through the prediction of a single site of disconnection at the molecule and the removal and/or addition of fragments at that site. A pipeline of multiple, identical Modof models is implemented into Modof-pipe to modify an input molecule at multiple disconnection sites. Here we show that Modof-pipe is able to retain major molecular scaffolds, allow controls over intermediate optimization steps and better constrain molecule similarities. Modof-pipe outperforms the state-of-the-art methods on benchmark datasets. Without molecular similarity constraints, Modof-pipe achieves 81.2% improvement in the octanol–water partition coefficient, penalized by synthetic accessibility and ring size, and 51.2%, 25.6% and 9.2% improvement if the optimized molecules are at least 0.2, 0.4 and 0.6 similar to those before optimization, respectively. Modof-pipe is further enhanced into Modof-pipem to allow modification of one molecule to multiple optimized ones. Modof-pipem achieves additional performance improvement, at least 17.8% better than Modof-pipe.

Ziqi Chen, Martin Renqiang Min, Srinivasan Parthasarathy, and Xia Ning. A deep generative model for molecule optimization via one fragment modification. Nature Machine Intelligence, 3:1040–1049, Dec. 2021. (Preprint is available here. Code is available here).

G2Retro as a Two-step Graph Generative Models for Retrosynthesis Prediction

a. G2Retro reaction center identification. G2Retro uses a graph message passing network (GMPN); G2Retro predicts three types of reaction centers: newly formed bonds (BF-center), bonds with type changes (BC-center), and atoms with leaving fragments (A-center); for BF-center, G2Retro also predicts bonds that have type changes induced by the newly formed bonds (BTCP); for all the reaction center types, G2Retro predicts atoms with charge changes (ACP). b. G2Retro synthon completion. G2Retro uses GMPN to represent both the products and the synthons; G2Retro sequentially predicts whether a new substructure should be attached (AACP) and the type of the attachment (AATP); G2Retro adds predicted substructures until AACP predicts ‘stop’.

Abstract: Retrosynthesis is a procedure where a target molecule is transformed into potential reactants and thus the synthesis routes can be identified. Recently, computational approaches have been developed to accelerate the design of synthesis routes. In this paper,we develop a generative framework G2Retro for one-step retrosynthesis prediction. G2Retro imitates the reversed logic of synthetic reactions. It first predicts the reaction centers in the target molecules (products), identifies the synthons needed to assemble the products, and transforms these synthons into reactants. G2Retro defines a comprehensive set of reaction center types, and learns from the molecular graphs of the products to predict potential reaction centers. To complete synthons into reactants, G2Retro considers all the involved synthon structures and the product structures to identify the optimal completion paths, and accordingly attaches small substructures sequentially to the synthons. Here we show that G2Retro is able to better predict the reactants for given products in the benchmark dataset than the state-of-the-art methods.

Ziqi Chen, Oluwatosin R. Ayinde, James R. Fuchs, Huan Sun, and Xia Ning. G2Retro: Two-step graph generative models for retrosynthesis prediction. Communications Chemistry, 6(1):102, 2023. (Preprint is available here. Code is available here. Webport is here.

Shape-conditioned 3D Molecule Generation via Equivariant Diffusion Models

Abstract: Ligand-based drug design aims to identify novel drug candidates of similar shapes with known active molecules. In this paper, we formulated an in silico shape-conditioned molecule generation problem to generate 3D molecule structures conditioned on the shape of a given molecule. To address this problem, we developed a translation- and rotation-equivariant shape-guided generative model ShapeMol. ShapeMol consists of an equivariant shape encoder that maps molecular surface shapes into latent embeddings, and an equivariant diffusion model that generates 3D molecules based on these embeddings. Experimental results show that ShapeMol can generate novel, diverse, drug-like molecules that retain 3D molecular shapes similar to the given shape condition. These results demonstrate the potential of ShapeMol in designing drug candidates of desired 3D shapes binding to protein target pockets.

Ziqi Chen, Bo Peng, Srinivasan Parthasarathy, and Xia Ning. Shape-conditioned 3D molecule generation via equivariant diffusion models. arXiv:2308.11890, August 2023.

RLSynC: Offline-Online Reinforcement Learning for Synthon Completion

Abstract: Retrosynthesis is the process of determining the set of reactant molecules that can react to form a desired product. Semi-template-based retrosynthesis methods, which im- itate the reverse logic of synthesis reactions, first predict the reaction centers in the products, and then complete the resulting synthons back into reactants. These methods enable necessary interpretability and high practical utility to inform synthesis planning. We develop a new offline-online reinforcement learning method RLSynC for synthon completion in semi-template-based methods. RLSynC assigns one agent to each synthon, all of which complete the synthons by conducting actions step by step in a synchronized fashion. RLSynC learns the policy from both offline training episodes and online interactions which allow RLSynC to explore new reaction spaces. RLSynC uses a forward synthesis model to evaluate the likelihood of the predicted reactants in synthesizing a product, and thus guides the action search. We compare RLSynC with the state-of-the-art retrosynthesis methods. Our experimental results demonstrate that RLSynC can outperform these methods with improvement as high as 14.9% on synthon completion, and 14.0% on retrosynthesis, highlighting its potential in synthesis planning.

Frazier Baker, Ziqi Chen, and Xia Ning. RLSynC: Offline-online reinforcement learning for synthon completion. page arXiv:2309.02671, September 2023.

Binding Peptide Generation for MHC Glass I Proteins with Deep Reinforcement Learning

Abstract: MHC Class I protein plays an important role in immunotherapy by presenting immunogenic peptides to anti-tumor immune cells. The repertoires of peptides for various MHC Class I proteins are distinct, which can be reflected by their diverse binding motifs. To characterize binding motifs for MHC Class I proteins, in vitro experiments have been conducted to screen peptides with high binding affinities to hundreds of given MHC Class I proteins. However, considering tens of thousands of known MHC Class I proteins, conducting in vitro experiments for extensive MHC proteins is infeasible, and thus a more efficient and scalable way to characterize binding motifs is needed. We presented a de novo generation framework, coined PepPPO, to characterize binding motif for any given MHC Class I proteins via generating repertoires of peptides presented by them. PepPPO leverages a reinforcement learning agent with a mutation policy to mutate random input peptides into positive presented ones. Using PepPPO, we characterized binding motifs for around 10 000 known human MHC Class I proteins with and without experimental data. These computed motifs demonstrated high similarities with those derived from experimental data. In addition, we found that the motifs could be used for the rapid screening of neoantigens at a much lower time cost than previous deep-learning methods.

Ziqi Chen, Baoyi Zhang, Hongyu Guo, Prashant Emani, Chongming Jiang, Mark Gerstein, Xia Ning, Chao Cheng, and Martin Renqiang Min. Binding peptide generation for MHC class I proteins with deep reinforcement learning. Bioinformatics, 39(2):btad055, 01 2023.