Proteome-Wide Profiling of the Covalent-Druggable Cysteines with a Structure-Based Deep Graph Learning Network

Machine Learning Models to Interrogate Proteome-Wide Covalent Ligandabilities Directed at Cysteines

Machine learning (ML) identification of covalently ligandable sites may accelerate targeted covalent inhibitor design and help expand the druggable proteome space. Here, we report the rigorous development and validation of the tree-based models and convolutional neural networks (CNNs) trained on a newly curated database (LigCys3D) of over 1000 liganded cysteines in nearly 800 proteins represented by over 10,000 three-dimensional structures in the protein data bank. The unseen tests yielded 94 and 93% area under the receiver operating characteristic curves for the tree models and CNNs, respectively. Based on the AlphaFold2 predicted structures, the ML models recapitulated the newly liganded cysteines in the PDB with over 90% recall values. To assist the community of covalent drug discoveries, we report the predicted ligandable cysteines in 392 human kinases and their locations in the sequence-aligned kinase structure, including the PH and SH2 domains. Furthermore, we disseminate a searchable online database LigCys3D (https://ligcys.computchem.org/) and a web prediction server DeepCys (https://deepcys.computchem.org/), both of which will be continuously updated and improved by including newly published experimental data. The present work represents an important step toward the ML-led integration of big genome data and structure models to annotate the human proteome space for the next-generation covalent drug discoveries.

Is fragment-based graph a better graph-based molecular representation for drug design? A comparison study of graph-based models

Graph Neural Networks (GNNs) have gained significant traction in various sectors of AI-driven drug design. Over recent years, the integration of fragmentation concepts into GNNs has emerged as a potent strategy to augment the efficacy of molecular generative models. Nonetheless, challenges such as symmetry breaking and potential misrepresentation of intricate cycles and undefined functional groups raise questions about the superiority of fragment-based graph representation over traditional methods. In our research, we undertook a rigorous evaluation, contrasting the predictive prowess of eight models-developed using deep learning algorithms-across 12 benchmark datasets that span a range of properties. These models encompass established methods like GCN, AttentiveFP, and D-MPNN, as well as innovative fragment-based representation techniques. Our results indicate that fragment-based methodologies, notably PharmHGT, significantly improve model performance and interpretability, particularly in scenarios characterized by limited data availability. However, in situations with extensive training, fragment-based molecular graph representations may not necessarily eclipse traditional methods. In summation, we posit that the integration of fragmentation, as an avant-garde technique in drug design, harbors considerable promise for the future of AI-enhanced drug design.

Machine Learning Models to Interrogate Proteomewide Covalent Ligandabilities Directed at Cysteines

Machine learning (ML) identification of covalently ligandable sites may accelerate targeted covalent inhibitor design and help expand the druggable proteome space. Here we report the rigorous development and validation of the tree-based models and convolutional neural networks (CNNs) trained on a newly curated database (LigCys3D) of over 1,000 liganded cysteines in nearly 800 proteins represented by over 10,000 three-dimensional structures in the protein data bank. The unseen tests yielded 94% and 93% AUCs (area under the receiver operating characteristic curve) for the tree models and CNNs, respectively. Based on the AlphaFold2 predicted structures, the ML models recapitulated the newly liganded cysteines in the PDB with over 90% recall values. To assist the community of covalent drug discoveries, we report the predicted ligandable cysteines in 392 human kinases and their locations in the sequence-aligned kinase structure including the PH and SH2 domains. Furthermore, we disseminate a searchable online database LigCys3D (https://ligcys.computchem.org/) and a web prediction server DeepCys (https://deepcys.computchem.org/), both of which will be continuously updated and improved by including newly published experimental data. The present work represents a first step towards the ML-led integration of big genome data and structure models to annotate the human proteome space for the next-generation covalent drug discoveries.

A flexible data-free framework for structure-based de novo drug design with reinforcement learning

Contemporary structure-based molecular generative methods have demonstrated their potential to model the geometric and energetic complementarity between ligands and receptors, thereby facilitating the design of molecules with favorable binding affinity and target specificity. Despite the introduction of deep generative models for molecular generation, the atom-wise generation paradigm that partially contradicts chemical intuition limits the validity and synthetic accessibility of the generated molecules. Additionally, the dependence of deep learning models on large-scale structural data has hindered their adaptability across different targets. To overcome these challenges, we present a novel search-based framework, 3D-MCTS, for structure-based de novo drug design. Distinct from prevailing atom-centric methods, 3D-MCTS employs a fragment-based molecular editing strategy. The fragments decomposed from small-molecule drugs are recombined under predefined retrosynthetic rules, offering improved drug-likeness and synthesizability, overcoming the inherent limitations of atom-based approaches. Leveraging multi-threaded parallel simulations combined with a real-time energy constraint-based pruning strategy, 3D-MCTS achieves remarkable efficiency. At a fixed computational cost, it outperforms other state-of-the-art (SOTA) methods by producing molecules with enhanced binding affinity. Furthermore, its fragment-based approach ensures the generation of more dependable binding conformations, exhibiting a success rate 43.6% higher than that of other SOTAs. This advantage becomes even more pronounced when handling targets that significantly deviate from the training dataset. 3D-MCTS is capable of achieving thirty times more hits with high binding affinity than traditional virtual screening methods, which demonstrates the superior ability of 3D-MCTS to explore chemical space. Moreover, the flexibility of our framework makes it easy to incorporate domain knowledge during the process, thereby enabling the generation of molecules with desirable pharmacophores and enhanced binding affinity. The adaptability of 3D-MCTS is further showcased in metalloprotein applications, highlighting its potential across various drug design scenarios.

Proteome-wide structural analysis identifies warhead- and coverage-specific biases in cysteine-focused chemoproteomics

Covalent drug discovery has undergone a resurgence over the past two decades and reactive cysteine profiling has emerged in parallel as a platform for ligand discovery through on- and off-target profiling; however, the scope of this approach has not been fully explored at the whole-proteome level. We combined AlphaFold2-predicted side-chain accessibilities for >95% of the human proteome with a meta-analysis of eighteen public cysteine profiling datasets, totaling 44,187 unique cysteine residues, revealing accessibility biases in sampled cysteines primarily dictated by warhead chemistry. Analysis of >3.5 million cysteine-fragment interactions further showed that hit elaboration and optimization drives increased bias against buried cysteine residues. Based on these data, we suggest that current profiling approaches cover a small proportion of potential ligandable cysteine residues and propose future directions for increasing coverage, focusing on high-priority residues and depth. All analysis and produced resources are freely available and extendable to other reactive amino acids.

AP-2δ Is the Most Relevant Target of AP-2 Family-Focused Cancer Therapy and Affects Genome Organization

Formerly hailed as "undruggable" proteins, transcription factors (TFs) are now under investigation for targeted therapy. In cancer, this may alter, inter alia, immune evasion or replicative immortality, which are implicated in genome organization, a process that accompanies multi-step tumorigenesis and which frequently develops in a non-random manner. Still, targeting-related research on some TFs is scarce, e.g., among AP-2 proteins, which are known for their altered functionality in cancer and prognostic importance. Using public repositories, bioinformatics tools, and RNA-seq data, the present study examined the ligandability of all AP-2 members, selecting the best one, which was investigated in terms of mutations, targets, co-activators, correlated genes, and impact on genome organization. AP-2 proteins were found to have the conserved "TF_AP-2" domain, but manifested different binding characteristics and evolution. Among them, AP-2δ has not only the highest number of post-translational modifications and extended strands but also contains a specific histidine-rich region and cleft that can receive a ligand. Uterine, colon, lung, and stomach tumors are most susceptible to AP-2δ mutations, which also co-depend with cancer hallmark genes and drug targets. Considering AP-2δ targets, some of them were located proximally in the spatial genome or served as co-factors of the genes regulated by AP-2δ. Correlation and functional analyses suggested that AP-2δ affects various processes, including genome organization, via its targets; this has been eventually verified in lung adenocarcinoma using expression and immunohistochemistry data of chromosomal conformation-related genes. In conclusion, AP-2δ affects chromosomal conformation and is the most appropriate target for cancer therapy focused on the AP-2 family.