Membrane Permeating Macrocycles: Design Guidelines from Machine Learning

Cyclic peptide membrane permeability prediction using deep learning model based on molecular attention transformer

Membrane permeability is a critical bottleneck in the development of cyclic peptide drugs. Experimental membrane permeability testing is costly, and precise in silico prediction tools are scarce. In this study, we developed CPMP (https://github.com/panda1103/CPMP), a cyclic peptide membrane permeability prediction model based on the Molecular Attention Transformer (MAT) frame. The model demonstrated robust predictive performance, achieving determination coefficients (R 2 ) of 0.67 for PAMPA permeability prediction, and R 2 values of 0.75, 0.62, and 0.73 for Caco-2, RRCK, and MDCK cell permeability predictions, respectively. Its performance outperforms traditional machine learning methods and graph-based neural network models. In ablation experiments, we validated the effectiveness of each component in the MAT architecture. Additionally, we analyzed the impact of data pre-training and cyclic peptide conformation optimization on model performance.

Meta-Analysis of Permeability Literature Data Shows Possibilities and Limitations of Popular Methods

Permeability is an important molecular property in drug discovery, as it co-determines pharmacokinetics whenever a drug crosses the phospholipid bilayer, e.g., into the cell, in the gastrointestinal tract, or across the blood-brain barrier. Many methods for the determination of permeability have been developed, including cell line assays (CACO-2 and MDCK), cell-free model systems like parallel artificial membrane permeability assay (PAMPA) mimicking, e.g., gastrointestinal epithelia or the skin, as well as the black lipid membrane (BLM) and submicrometer liposomes. Furthermore, many in silico approaches have been developed for permeability prediction: meta-analysis of publicly available databases for permeability data (MolMeDB and ChEMBL) was performed to establish their usability. Four experimental and two computational methods were evaluated. It was shown that repeatability of the reported permeability measurement is not great even for the same method. For the PAMPA method, two different permeabilities are reported: intrinsic and apparent. They can vary in degrees of magnitude; thus, we suggest being extra cautious using literature data on permeability. When we compared data for the same molecules using different methods, the best agreement was between cell-based methods and between BLM and computational methods. Existence of unstirred water layer (UWL) permeability limits the data agreement between cell-based methods (and apparent PAMPA) with data that are not limited by UWL permeability (computational methods, BLM, intrinsic PAMPA). Therefore, different methods have different limitations. Cell-based methods provide results only in a small range of permeabilities (-8 to -4 in cm/s), and computational methods can predict a wider range of permeabilities beyond physical limitations, but their precision is therefore limited. BLM with liposomes can be used for both fast and slow permeating molecules, but its usage is more complicated than standard transwell techniques. To sum up, when working with in-house measured or published permeability data, we recommend caution in interpreting and combining them.

A membrane permeability database for nonpeptidic macrocycles

The process of developing new drugs is arduous and costly, particularly for targets classified as "difficult-to-drug." Macrocycles show a particular ability to modulate difficult-to-drug targets, including protein-protein interactions, while still allowing oral administration. However, the determination of membrane permeability, critical for reaching intracellular targets and for oral bioavailability, is laborious and expensive. In silico methods are a cost-effective alternative, enabling predictions prior to compound synthesis. Here, we present a comprehensive online database ( https://swemacrocycledb.com/ ), housing 5638 membrane permeability datapoints for 4216 nonpeptidic macrocycles, curated from the literature, patents, and bioactivity repositories. In addition, we present a new descriptor, the "amide ratio" (AR), that quantifies the peptidic nature of macrocyclic compounds, enabling the classification of peptidic, semipeptidic, and nonpeptidic macrocycles. Overall, this resource fills a gap among existing databases, offering valuable insights into the membrane permeability of nonpeptidic and semipeptidic macrocycles, and facilitating predictions for drug discovery projects.

Multi_CycGT: A Deep Learning-Based Multimodal Model for Predicting the Membrane Permeability of Cyclic Peptides

Cyclic peptides are gaining attention for their strong binding affinity, low toxicity, and ability to target "undruggable" proteins; however, their therapeutic potential against intracellular targets is constrained by their limited membrane permeability, and researchers need much time and money to test this property in the laboratory. Herein, we propose an innovative multimodal model called Multi_CycGT, which combines a graph convolutional network (GCN) and a transformer to extract one- and two-dimensional features for predicting cyclic peptide permeability. The extensive benchmarking experiments show that our Multi_CycGT model can attain state-of-the-art performance, with an average accuracy of 0.8206 and an area under the curve of 0.8650, and demonstrates satisfactory generalization ability on several external data sets. To the best of our knowledge, it is the first deep learning-based attempt to predict the membrane permeability of cyclic peptides, which is beneficial in accelerating the design of cyclic peptide active drugs in medicinal chemistry and chemical biology applications.

Current advances and development strategies of orally bioavailable PROTACs

Proteolysis-targeting chimeras (PROTACs) have been an area of intensive research with the potential to extend drug space not target to traditional molecules. In the last half decade, we have witnessed several PROTACs initiated phase I/II/III clinical trials, which inspired us a lot. However, the structure of PROTACs beyond "rule of 5" resulted in developing PROTACs with acceptable oral pharmacokinetic (PK) properties remain one of the biggest bottleneck tasks. Many reports have demonstrated that it is possible to access orally bioavailable PROTACs through rational ligand and linker modifications. In this review, we systematically reviewed and highlighted the most recent advances in orally bioavailable PROTACs development, especially focused on the medicinal chemistry campaign of discovery process and in vivo oral PK properties. Moreover, the constructive strategies for developing oral PROTACs were proposed comprehensively. Collectively, we believe that the strategies summarized here may provide references for further development of oral PROTACs.

Addressing Challenges of Macrocyclic Conformational Sampling in Polar and Apolar Solvents: Lessons for Chameleonicity

We evaluated a workflow to reliably sample the conformational space of a set of 47 peptidic macrocycles. Starting from SMILES strings, we use accelerated molecular dynamics simulations to overcome high energy barriers, in particular, the cis-trans isomerization of peptide bonds. We find that our approach performs very well in polar solvents like water and dimethyl sulfoxide. Interestingly, the protonation state of a secondary amine in the ring only slightly influences the conformational ensembles of our test systems. For several of the macrocycles, determining the conformational distribution in chloroform turns out to be considerably more challenging. Especially, the choice of partial charges crucially influences the ensembles in chloroform. We address these challenges by modifying initial structures and the choice of partial charges. Our results suggest that special care has to be taken to understand the configurational distribution in apolar solvents, which is a key step toward a reliable prediction of membrane permeation of macrocycles and their chameleonic properties.

Adaptation of Empirical Methods to Predict the LogD of Triazine Macrocycles

Octanol/water partition coefficients guide drug design, but algorithms do not always accurately predict these values. For cationic triazine macrocycles that adopt a conserved folded shape in solution, common algorithms fall short. Here, the logD values for 12 macrocycles differing in amino acid choice were predicted and then measured experimentally. On average, AlogP, XlogP, and ChemAxon predictions deviate by 0.9, 2.8, and 3.9 log units, with XlogP overestimating lipophilicity and AlogP and ChemAxon underestimating lipophilicity. Importantly, however, a linear relationship (R2 > 0.98) exists between the values predicted by AlogP and the experimentally determined logD values, thus enabling more accurate predictions.

Combining SILCS and Artificial Intelligence for High-Throughput Prediction of the Passive Permeability of Drug Molecules

Membrane permeability of drug molecules plays a significant role in the development of new therapeutic agents. Accordingly, methods to predict the passive permeability of drug candidates during a medicinal chemistry campaign offer the potential to accelerate the drug design process. In this work, we combine the physics-based site identification by ligand competitive saturation (SILCS) method and data-driven artificial intelligence (AI) to create a high-throughput predictive model for the passive permeability of druglike molecules. In this study, we present a comparative analysis of four regression models to predict membrane permeabilities of small druglike molecules; of the tested models, Random Forest was the most predictive yielding an R2 of 0.81 for the independent data set. The input feature vector used to train the developed prediction model includes absolute free energy profiles of ligands through a POPC-cholesterol bilayer based on ligand grid free energy (LGFE) profiles obtained from the SILCS approach. The use of the membrane free energy profiles from SILCS offers information on the physical forces contributing to ligand permeability, while the use of AI yields a more predictive model trained on experimental PAMPA permeability data for a collection of 229 molecules. This combination allows for rapid estimations of ligand permeability at a level of accuracy beyond currently available predictive models while offering insights into the contributions of the functional groups in the ligands to the permeability barrier, thereby offering quantitative information to facilitate rational ligand design.

Targeted Protein Degradation: Advances, Challenges, and Prospects for Computational Methods

The therapeutic approach of targeted protein degradation (TPD) is gaining momentum due to its potentially superior effects compared with protein inhibition. Recent advancements in the biotech and pharmaceutical sectors have led to the development of compounds that are currently in human trials, with some showing promising clinical results. However, the use of computational tools in TPD is still limited, as it has distinct characteristics compared with traditional computational drug design methods. TPD involves creating a ternary structure (protein-degrader-ligase) responsible for the biological function, such as ubiquitination and subsequent proteasomal degradation, which depends on the spatial orientation of the protein of interest (POI) relative to E2-loaded ubiquitin. Modeling this structure necessitates a unique blend of tools initially developed for small molecules (e.g., docking) and biologics (e.g., protein-protein interaction modeling). Additionally, degrader molecules, particularly heterobifunctional degraders, are generally larger than conventional small molecule drugs, leading to challenges in determining drug-like properties like solubility and permeability. Furthermore, the catalytic nature of TPD makes occupancy-based modeling insufficient. TPD consists of multiple interconnected yet distinct steps, such as POI binding, E3 ligase binding, ternary structure interactions, ubiquitination, and degradation, along with traditional small molecule properties. A comprehensive set of tools is needed to address the dynamic nature of the induced proximity ternary complex and its implications for ubiquitination. In this Perspective, we discuss the current state of computational tools for TPD. We start by describing the series of steps involved in the degradation process and the experimental methods used to characterize them. Then, we delve into a detailed analysis of the computational tools employed in TPD. We also present an integrative approach that has proven successful for degrader design and its impact on project decisions. Finally, we examine the future prospects of computational methods in TPD and the areas with the greatest potential for impact.

Lessons for Oral Bioavailability: How Conformationally Flexible Cyclic Peptides Enter and Cross Lipid Membranes

Cyclic peptides extend the druggable target space due to their size, flexibility, and hydrogen-bonding capacity. However, these properties impact also their passive membrane permeability. As the "journey" through membranes cannot be monitored experimentally, little is known about the underlying process, which hinders rational design. Here, we use molecular simulations to uncover how cyclic peptides permeate a membrane. We show that side chains can act as "molecular anchors", establishing the first contact with the membrane and enabling insertion. Once inside, the peptides are positioned between headgroups and lipid tails─a unique polar/apolar interface. Only one of two distinct orientations at this interface allows for the formation of the permeable "closed" conformation. In the closed conformation, the peptide crosses to the lower leaflet via another "anchoring" and flipping mechanism. Our findings provide atomistic insights into the permeation process of flexible cyclic peptides and reveal design considerations for each step of the process.