Predictive machine learning models for rational permeability design in de novo macrocycle engineering: a review.

Small molecule drug discovery has been highly successful across many therapeutic areas over decades of progress; however, many disease-relevant proteins remain difficult to target. In particular, intracellular proteins with large, shallow, or flexible interaction surfaces are poorly addressed by classical drug-like compounds. For these reasons, drug discovery efforts have shifted toward alternative molecular classes. This has led to growing interest in macrocyclic compounds in recent years, which have emerged as an important class of therapeutic molecules, particularly for targets that are out of reach for conventional small molecules. These compounds operate within the chemical space beyond Lipinski's Rule of Five (bRo5) and offer new opportunities for modulating difficult intracellular targets. At the same time, their size, flexibility, and structural complexity introduce significant challenges, among which the accurate prediction of membrane permeability remains one of the persistent limitations in their rational design, particularly for orally bioavailable candidates. This review examines 11 published studies, specifically those focusing on machine learning (ML) and deep learning (DL) driven methodologies for predicting macrocycle permeability. Contemporary ML models utilize a diverse array of molecular representations, including two-dimensional descriptors, three-dimensional conformational features, molecular graphs, and Simplified Molecular Input Line Entry System (SMILES) sequences, all aimed at accelerating prediction pipelines. The architectural range of these models spans from classical Quantitative Structure-Permeability Relationship (QSPR) algorithms to advanced deep learning frameworks, including Transformer-based models, Graph Neural Networks (GNNs), and vision-based approaches. Significant progress in this domain has been largely powered by the establishment of the CycPeptMPDB (Cyclic Peptide Membrane Permeability Database), which provides standardized permeability annotations for thousands of cyclic peptides and has facilitated progress in predictive modeling. Nevertheless, several substantial challenges remain, such as issues of poor generalizability across different chemical scaffolds, the inherent limitations in interpretability characteristic of deep learning models, and biases frequently introduced by dataset imbalance or the particular details of permeability assays. Non-peptidic macrocycles are of growing relevance in drug discovery, but there are remarkably few proposed models that have been developed for this particular subclass. The recent introduction of NPMMPD, a curated dataset focusing on non-peptidic scaffolds offers a promising path toward further advances. This database will allow the development of ML models for the early-stage screening and rational design of orally available non-peptidic macrocycles. Ultimately, these advancements highlight the critical importance of integrating mechanistic biophysical insights with data-driven learning paradigms to accelerate the discovery of the oral therapeutic potential of macrocycles within the bRo5 chemical space.