Perplexity-Based Molecule Ranking and Bias Estimation of Chemical Language Models

Deep learning for low-data drug discovery: Hurdles and opportunities

Deep learning is becoming increasingly relevant in drug discovery, from de novo design to protein structure prediction and synthesis planning. However, it is often challenged by the small data regimes typical of certain drug discovery tasks. In such scenarios, deep learning approaches-which are notoriously 'data-hungry'-might fail to live up to their promise. Developing novel approaches to leverage the power of deep learning in low-data scenarios is sparking great attention, and future developments are expected to propel the field further. This mini-review provides an overview of recent low-data-learning approaches in drug discovery, analyzing their hurdles and advantages. Finally, we venture to provide a forecast of future research directions in low-data learning for drug discovery.

Exploring data-driven chemical SMILES tokenization approaches to identify key protein-ligand binding moieties

Machine learning models have found numerous successful applications in computational drug discovery. A large body of these models represents molecules as sequences since molecular sequences are easily available, simple, and informative. The sequence-based models often segment molecular sequences into pieces called chemical words, analogous to the words that make up sentences in human languages, and then apply advanced natural language processing techniques for tasks such as de novo drug design, property prediction, and binding affinity prediction. However, the chemical characteristics and significance of these building blocks, chemical words, remain unexplored. To address this gap, we employ data-driven SMILES tokenization techniques such as Byte Pair Encoding, WordPiece, and Unigram to identify chemical words and compare the resulting vocabularies. To understand the chemical significance of these words, we build a language-inspired pipeline that treats high affinity ligands of protein targets as documents and selects key chemical words making up those ligands based on tf-idf weighting. The experiments on multiple protein-ligand affinity datasets show that despite differences in words, lengths, and validity among the vocabularies generated by different subword tokenization algorithms, the identified key chemical words exhibit similarity. Further, we conduct case studies on a number of target to analyze the impact of key chemical words on binding. We find that these key chemical words are specific to protein targets and correspond to known pharmacophores and functional groups. Our approach elucidates chemical properties of the words identified by machine learning models and can be used in drug discovery studies to determine significant chemical moieties.

Using alternative SMILES representations to identify novel functional analogues in chemical similarity vector searches

Chemical similarity searches are a widely used family of in silico methods for identifying pharmaceutical leads. These methods historically relied on structure-based comparisons to compute similarity. Here, we use a chemical language model to create a vector-based chemical search. We extend previous implementations by creating a prompt engineering strategy that utilizes two different chemical string representation algorithms: one for the query and the other for the database. We explore this method by reviewing search results from nine queries with diverse targets. We find that the method identifies molecules with similar patent-derived functionality to the query, as determined by our validated LLM-assisted patent summarization pipeline. Further, many of these functionally similar molecules have different structures and scaffolds from the query, making them unlikely to be found with traditional chemical similarity searches. This method may serve as a new tool for the discovery of novel molecular structural classes that achieve target functionality.

A Simple Way to Incorporate Target Structural Information in Molecular Generative Models

Deep learning generative models are now being applied in various fields including drug discovery. In this work, we propose a novel approach to include target 3D structural information in molecular generative models for structure-based drug design. The method combines a message-passing neural network model that predicts docking scores with a generative neural network model as its reward function to navigate the chemical space searching for molecules that bind favorably with a specific target. A key feature of the method is the construction of target-specific molecular sets for training, designed to overcome potential transferability issues of surrogate docking models through a two-round training process. Consequently, this enables accurate guided exploration of the chemical space without reliance on the collection of prior knowledge about active and inactive compounds for the specific target. Tests on eight target proteins showed a 100-fold increase in hit generation compared to conventional docking calculations and the ability to generate molecules similar to approved drugs or known active ligands for specific targets without prior knowledge. This method provides a general and highly efficient solution for structure-based molecular generation.

Can We Quickly Learn to "Translate" Bioactive Molecules with Transformer Models?

Meaningful exploration of the chemical space of druglike molecules in drug design is a highly challenging task due to a combinatorial explosion of possible modifications of molecules. In this work, we address this problem with transformer models, a type of machine learning (ML) model originally developed for machine translation. By training transformer models on pairs of similar bioactive molecules from the public ChEMBL data set, we enable them to learn medicinal-chemistry-meaningful, context-dependent transformations of molecules, including those absent from the training set. By retrospective analysis on the performance of transformer models on ChEMBL subsets of ligands binding to COX2, DRD2, or HERG protein targets, we demonstrate that the models can generate structures identical or highly similar to most active ligands, despite the models having not seen any ligands active against the corresponding protein target during training. Our work demonstrates that human experts working on hit expansion in drug design can easily and quickly employ transformer models, originally developed to translate texts from one natural language to another, to "translate" from known molecules active against a given protein target to novel molecules active against the same target.

Exposing the Limitations of Molecular Machine Learning with Activity Cliffs

Machine learning has become a crucial tool in drug discovery and chemistry at large, e.g., to predict molecular properties, such as bioactivity, with high accuracy. However, activity cliffs─pairs of molecules that are highly similar in their structure but exhibit large differences in potency─have received limited attention for their effect on model performance. Not only are these edge cases informative for molecule discovery and optimization but also models that are well equipped to accurately predict the potency of activity cliffs have increased potential for prospective applications. Our work aims to fill the current knowledge gap on best-practice machine learning methods in the presence of activity cliffs. We benchmarked a total of 24 machine and deep learning approaches on curated bioactivity data from 30 macromolecular targets for their performance on activity cliff compounds. While all methods struggled in the presence of activity cliffs, machine learning approaches based on molecular descriptors outperformed more complex deep learning methods. Our findings highlight large case-by-case differences in performance, advocating for (a) the inclusion of dedicated "activity-cliff-centered" metrics during model development and evaluation and (b) the development of novel algorithms to better predict the properties of activity cliffs. To this end, the methods, metrics, and results of this study have been encapsulated into an open-access benchmarking platform named MoleculeACE (Activity Cliff Estimation, available on GitHub at: https://github.com/molML/MoleculeACE). MoleculeACE is designed to steer the community toward addressing the pressing but overlooked limitation of molecular machine learning models posed by activity cliffs.

Naming of new natural products: Standard, pitfalls and tips-and-tricks

Naming a newly discovered natural product (NP) is a pleasant but difficult exercise. In most cases, the NP name will be given with reference to the species of origin, be it a plant, a marine organism, a mammalian or microbial species. For a long time, the use of biologically-based trivial names has been recommended to identify the parental linkage between the product and the originating genus or species. But the recommendation is not always followed and a multiplicity of trivial names have been attributed to NP, based on locations (country, region, city), foods, music, animals, forenames, etc. Tips-and-tricks associated with the naming of NP are underlined here. Usually, NP are differentiated across a homogeneous chemical series with a letter (from the Latin or Greek alphabet), followed or not with a number. In other cases, the change of a single letter distinguishes a series of NP. Common pitfalls associated with the naming of NP are enumerated, including the complexity of names, use of synonyms, duplicated names, confusing names and inappropriate terminology. The difficulties regularly encountered with the naming of NP are discussed. Four essential recommendations are recalled: (i) a thorough analysis of the existing products to avoid duplicated names and confusion, (ii) the use of a biologically-based trivial name to retrace the origin of the product, (iii) the strict adherence to the codes of chemical nomenclature, and (iv) the preference for simple names to facilitate transmission. Naming a new NP is a rewarding task, which shall be performed with all due skill, care and diligence.

DeepAS - Chemical language model for the extension of active analogue series

In medicinal chemistry, hit-to-lead and lead optimization efforts produce analogue series (ASs), the analysis of which is of central relevance for the exploration and exploitation of structure-activity relationships (SARs) and generation of candidate compounds. The key question in any chemical optimization effort is which analogue(s) to generate next, for which computational support is typically provided through QSAR analysis and compound potency predictions. In this study, we introduce a new chemical language model for analogue design via deep learning. For this purpose, ASs comprising active compounds are ordered according to increasing potency and the chemical language model predicts preferred R-groups for new analogues on the basis of ordered R-group sequences. Hence, consistent with the principles of deep models for natural language processing, analogues with new R-groups are predicted based upon conditional probabilities taking preceding groups into account. This implicitly accounts for the potency gradient captured by an AS and detectable SAR trends, providing a new concept for analogue design. Herein, we report the AS-based chemical language model, its initial evaluation, and exemplary applications.