Journal of Cheminformatics最新文献

Generative artificial intelligence (GenAI) models have emerged as a transformative tool for addressing the complex challenges of drug discovery, enabling the design of structurally diverse, chemically valid, and functionally relevant molecules. Despite significant advancements, the rapid expansion of GenAI applications still faces challenges related to prediction accuracy, molecular validity, and optimization for drug-like properties. This review provides a comprehensive analysis of recent techniques and strategies aimed at enhancing the performance of GenAI models in molecular design. We explore key generative architectures, including variational autoencoders, generative adversarial networks, and transformer-based models, highlighting their unique contributions to drug discovery. Additionally, we discuss critical advancements such as reinforcement learning, multi-objective optimization, and the integration of domain-specific chemical knowledge, which collectively enhance molecular validity, novelty, and drug-likeness. Also, the review examines persistent challenges, including data quality limitations, model interpretability, and the need for improved objective functions, while offering insights into future research directions. By mapping the evolving landscape of GenAI-driven molecular design and providing strategic guidance for overcoming existing limitations, this review serves as an essential resource for researchers leveraging GenAI in drug discovery.

Interactions of proteins with other molecules are often mediated by a set of critical binding motifs on their surfaces. Most traditional binder designs relied on motifs borrowed from known binder molecules, which highly restricted their applicability to novel targets or new binding sites. This work presents a deep learning network MotifGen that predicts potential binder motifs directly from receptor structures without further supporting information. MotifGen generates motif profiles at the receptor surface for 14 types of functional groups or 6 chemical interaction classes. These profiles are highly human-interpretable and can be further utilized as pre-trained embedding inputs for versatile few-shot binder design applications. We demonstrate MotifGen's effectiveness through its applications to peptide binder design and small molecule binding site prediction, where it either surpassed existing methods or added significant value when integrated. Our motif-centric approach can offer a new design strategy for novel binder discovery for challenging receptor targets.

Methods for automatic chemical retrosynthesis have found recent success through the application of models traditionally built for natural language processing, primarily through transformer neural networks. These models have demonstrated significant ability to translate between the SMILES encodings of chemical products and reactants, but are constrained as a result of their autoregressive nature. We propose (texttt {DiffER}), an alternative template-free method for single-step retrosynthesis prediction in the form of categorical diffusion, which allows the entire output SMILES sequence to be predicted in unison. We construct an ensemble of diffusion models which achieves state-of-the-art performance for top-1 accuracy and competitive performance for top-3, top-5, and top-10 accuracy among template-free methods. We prove that (texttt {DiffER}) is a strong baseline for a new class of template-free model and is capable of learning a variety of synthetic techniques used in laboratory settings.

The human Ether-à-go-go-Related Gene (hERG) potassium channel is crucial for repolarizing the cardiac action potential and regulating the heartbeat. Molecules that inhibit this protein can cause acquired long QT syndrome, increasing the risk of arrhythmias and sudden fatal cardiac arrests. Detecting compounds with potential hERG inhibitory activity is therefore essential to mitigate cardiotoxicity risks. In this article, we present a new hERG data set of unprecedented size, comprising nearly 300,000 molecules reported in PubChem and ChEMBL, approximately 2000 of which were confirmed hERG blockers identified through in vitro assays. Multiple structure-based artificial intelligence (AI) binary classifiers for predicting hERG inhibitors were developed, employing, as descriptors, protein–ligand extended connectivity (PLEC) fingerprints fed into random forest, extreme gradient boosting, and deep neural network (DNN) algorithms. Our best-performing model, a stacking ensemble classifier with a DNN meta-learner, achieved state-of-the-art classification performance, accurately identifying 86% of molecules having half-maximal inhibitory concentrations (IC₅₀s) not exceeding 20 µM in our challenging test set, including 94% of hERG blockers whose IC₅₀s were not greater than 1 µM. It also demonstrated superior screening power compared to virtual screening schemes that used existing scoring functions. This model, named “HERGAI,” along with relevant input/output data and user-friendly source code, is available in our GitHub repository (https://github.com/vktrannguyen/HERGAI) and can be used to predict drug-induced hERG blockade, even on large data sets.

Advancements in cheminformatics have led to numerous methods for encoding molecules numerically. The choice of molecular representation impacts the accuracy and generalizability of learning algorithms applied to chemical datasets. Designing and selecting the appropriate representation often lacks a systematic approach and follows computationally exhaustive empirical testing. Moreover, research has shown that deep learning models do not substantially outperform traditional approaches across many tasks with no clear explanation for this shortfall. In this work, we present TopoLearn, a model that predicts the effectiveness of representations on datasets based on the topological characteristics of the corresponding feature space. Using interpretability techniques, we find that persistent homology descriptors are linked with the error metrics of trained machine learning models, offering a new method to better understand and select molecular representations.

Scientific contribution Our research is the first to establish an empirical connection between the topology of feature spaces and the machine learning performance of molecular representations. In addition, we facilitate future research endeavors by providing open access to our developed model.

This study, focusing on predicting Absorption, Distribution, Metabolism, Excretion, and Toxicology (ADMET) properties, addresses the key challenges of ML models trained using ligand-based representations. We propose a structured approach to data feature selection, taking a step beyond the conventional practice of combining different representations without systematic reasoning. Additionally, we enhance model evaluation methods by integrating cross-validation with statistical hypothesis testing, adding a layer of reliability to the model assessments. Our final evaluations include a practical scenario, where models trained on one source of data are evaluated on a different one. This approach aims to bolster the reliability of ADMET predictions, providing more dependable and informative model evaluations.

Scientific contribution

This study provided a structured approach to feature selection. We improve model evaluation by combining cross-validation with statistical hypothesis testing, making results more reliable. The methodology used in our study can be generalized beyond feature selection, boosting the confidence in selected models which is crucial in a noisy domain such as the ADMET prediction tasks. Additionally, we assess how well models trained on one dataset perform on another, offering practical insights for using external data in drug discovery.