Adrian Whitty

SUMMARY: Protein–ligand interactions underpin biological regulation and drug action, with both binding affinity and binding kinetics shaping functional outcomes. By analysing kinetic data for 4,311 protein-small-molecule pairs, we find that when association occurs below the diffusion-controlled limit, the rates of ligand association ( k _on ) and dissociation ( k _off ) are primarily determined by how the initial encounter complex reorganizes into the final bound state, and that this reorganization is governed chiefly by the intrinsic dynamic properties of the protein rather than by structural features of the ligand. Counterintuitively, therefore, k _off exhibits minimal dependence on ligand structure, so that dissociation proceeds through protein-gated conformational transitions rather than through direct rupture of protein–ligand contacts. This mechanistic behaviour stands in marked contrast to that for protein–protein complexes, based on an analysis of 1,561 interactions. Together, these findings challenge prevailing assumptions regarding the molecular determinants of small-molecule binding kinetics, and have broad implications for rationally modulating protein–ligand interactions and drug-target residence times.

Several influential studies published around the turn of the millennium reported that advancing a lead compound to a drug tended to involve certain predictable changes in molecular properties, most notably marked increases in molecular weight (MW) and in lipophilicity. To assess how lead optimization practices have changed over the past 20 years, we examined oral drugs approved during 2020-2023 and their corresponding lead compounds with respect to their molecular properties. As well as comparing property averages, we analyzed the changes introduced in each individual drug discovery program to directly assess the trends that characterize the modern lead-to-drug process. We found that changes in MW from lead to drug are substantially smaller than in the older studies, with a median mass change corresponding to addition of just a single heavy atom, and cLogP is now slightly more likely to decrease than to become greater. Thus, the notion that lead optimization tends to involve making the compound ‘bigger and greasier’ no longer applies. Rather, approaches have been found to refine compound potency and selectivity, and to eliminate ADME liabilities, without adding substantial mass or overly relying on the hydrophobic effect as a driver of binding. One such tactic may be avoiding addition of aromatic groups during lead optimization, coupled with a tendency to favor heterocycles over aromatic carbocycles. Overall, our results imply that it may no longer be useful to think of ‘lead-like’ and ‘drug-like’ compounds as occupying distinct property space.

Protein arginine methyl transferase 5 (PRMT5) plays a global role in cell physiology and is an established therapeutic target in cancer. In approximately 10-15% of human cancers, deletion of the methylthioadenosine phosphorylase (MTAP) gene results in accumulation of methylthioadenosine (MTA), exposing a synthetic lethality and opportunity for precision medicine by selective targeting of PRMT5 in this context. Reported small molecule PRMT5 inhibitors engage either cosubstrate S-adenosyl methionine (SAM) or peptide-substrate pockets through diverse mechanisms. A subset of chemotypes demonstrate uncompetitive engagement with SAM or its inhibitory metabolic precursor, MTA. Although uncompetitive engagement can be evaluated in cell-free systems, no methods exist to directly assess this in cells. Here, we describe the development of a fluorescent probe that acts as a dynamic BRET biosensor of the intracellular SAM/MTA pool that overcomes the current limitations of competitive binding analyses. Using this biosensor, we evaluate a range of diverse PRMT5 inhibitors to mechanistically characterize and quantify uncompetitive target engagement as well as ternary complex formation at PRMT5-SAM and PRMT5-MTA complexes in live cells, enabling direct insights into drug mechanism-of-action and metabolite-dependent responses of inhibitors.

We have investigated the impact of conformational diversity on the prediction of druggability to see whether structural ensembles of protein targets offer more precise insights. The study is based on binding hot spot analyses performed on 37 binding sites from 33 proteins adapted from well-known druggability benchmark sets. Binding hot spots are regions on proteins that significantly influence binding free energy and hence can be used to predict druggability. Using fast Fourier transform-based algorithms, small organic probe molecules are docked to pinpoint these hot spots with denser probe clusters indicating higher binding affinity. The binding sites are mapped across the structural ensemble of protein crystal structures from the Protein Data Bank with 90% sequence identity. Druggability is analyzed according to the hot spot strength, connectivity, compactness, and maximum dimension. Our results show that a protein's druggability depends on consensus across the structural ensemble, requiring approximately 70% of structures to have a strong hot spot at the binding site of interest and approximately 50% of structures to meet all three druggability criteria. The ability to occasionally access a rare druggable conformation is not sufficient for a protein to be druggable in practice. Hot spot strength proves to be crucial for druggability. However, failing to meet the secondary druggability criteria of connectivity/compactness and maximum dimension for 30% or more structures indicates the inhomogeneity of the ensemble and signals the need for a more detailed analysis of mapping results. Such inhomogeneity may occur due to conformational differences caused by the binding of charged ligands or by the substantial flexibility of the binding site. The hot spots can be determined by the public server FTMove, and the codes for processing the server output for druggability analysis are available on GitHub.

Macrocycles are emerging as a prominent modality in drug discovery, including for conventionally druggable targets for which simpler, acyclic ligands are readily discoverable. Given the additional synthetic challenges associated with macrocyclic chemotypes, we address what benefits macrocycles provided for these highly druggable targets. To do this, we examine the effects of macrocyclization on inhibitors of highly druggable kinase targets. For each example, we isolate closely matched acyclic/macrocyclic compound pairs, allowing us to pinpoint the effects of macrocyclization on binding affinity, selectivity, and ADME properties absent confounding factors. Our findings show that while the impact of macrocyclization on potency is variable, a profound effect on selectivity is common. Macrocyclization can also bring benefits for membrane permeability, efflux ratio, blood-brain barrier penetrance, and metabolic stability. These findings lead us to propose specific circumstances in which a drug discoverer targeting kinases or other conventionally druggable target classes should consider a macrocycle approach.

Macrocycles are emerging as a prominent modality in drug discovery, including for conventionally druggable targets for which simpler, acyclic ligands are readily discoverable. Given the additional synthetic challenges associated with macrocyclic chemotypes, we address what benefits macrocycles provided for these highly druggable targets. To do this, we examine the effects of macrocyclization on inhibitors of highly druggable kinase targets. For each example, we isolate closely matched acyclic/macrocyclic compound pairs, allowing us to pinpoint the effects of macrocyclization on binding affinity, selectivity, and ADME properties absent confounding factors. Our findings show that, while the impact of macrocyclization on potency is variable, a profound effect on selectivity is common. Macrocyclization can also bring benefits for membrane permeability, efflux ratio, blood-brain barrier penetrance, and metabolic stability. These findings lead us to propose specific circumstances in which a drug discoverer targeting kinases or other conventionally druggable target classes should consider a macrocycle approach.

Machine learning (ML) is rapidly gaining traction in many areas of experimental molecular science for elucidating relationships and patterns in large or complex data sets. Historically, ML was largely the preserve of those with specialized training in fields such as statistics or cheminformatics. Increasingly, however, ML methodologies are becoming part of the standard toolkit for experimental scientists across a range of disciplines. For scientists without a significant background in computer science or statistics, lowering the barrier of entry to these ML techniques is important to broadening access to these powerful methods. Here we provide detailed, step‐by‐step protocols for performing four ML methods that are particularly useful for applications in biochemistry, cell biology, and drug discovery: hierarchical clustering, principal component analysis (PCA), partial least squares discriminant analysis (PLSDA), and partial least squares regression (PLSR). The protocols are written for the widely used software MATLAB, but no prior experience with MATLAB is required to use them. We include an explanation of each step, pitched at a level to be understood by investigators without any prior experience with ML, MATLAB, or any kind of coding. We also highlight the scientific issues pertaining to selecting and scaling the data to be analyzed. Throughout, we emphasize the relationship between the scientific question and how to choose data and methods that will allow it to be addressed in a meaningful way. Our aim is to provide a basic introduction that will equip experimental chemical biologists, chemists, and other biomedical scientists with the knowledge required to use ML to aid in the design of experiments, the formulation and data‐driven testing of hypotheses, and the analysis of experimental data. © 2025 Wiley Periodicals LLC.Basic Protocol 1: ClusteringBasic Protocol 2: Principal component analysisBasic Protocol 3: Partial least squares‐discriminant analysisBasic Protocol 4: Partial least squares regression

Machine learning (ML) is rapidly gaining traction in many areas of experimental molecular science for elucidating relationships and patterns in large or complex data sets. Historically, ML was largely the preserve of those with specialized training in fields such as statistics or cheminformatics. Increasingly, however, ML methodologies are becoming part of the standard toolkit for experimental scientists across a range of disciplines. Lowering the barrier of entry to these ML techniques, for scientists without a significant background in computer science or statistics, is important to broadening access to these powerful methods. Here we provide detailed, step by step tutorials for performing four ML methods that are particularly useful for applications in biochemistry, cell biology, and drug discovery: hierarchical clustering, Principal Component Analysis (PCA), Partial Least-Squares Discriminant Analysis (PLSDA), and Partial Least-Squares Regression (PLSR). The protocols are written for the widely used software MATLAB, but no prior experience with MATLAB is required to use them. We include an explanation of each step, pitched at a level to be understood by investigators without any prior experience with ML, MATLAB, or any kind of coding. We also highlight the scientific issues pertaining to selecting and scaling the data to be analyzed, and describe controls to test the validity of the results obtained. Throughout, we emphasize the relationship between the scientific question and how to choose data and methods that will allow it to be addressed in a meaningful way. Our aim is to provide a basic introduction that will equip experimental chemical biologists and other chemical and biomedical scientists with the knowledge required to use ML to aid in the design of experiments, the formulation and data-driven testing of hypotheses, and the analysis of experimental data. Basic Protocol 1: Clustering Basic Protocol 2: Principal Component Analysis (PCA) Basic Protocol 3: Partial Least Squares Discriminant Analysis (PLSDA) Basic Protocol 4: Partial Least Squares Regression (PLSR)

Enhancing protein thermal stability is important for biomedical and industrial applications as well as in the research laboratory. Here, we describe a simple machine-learning method which identifies amino acid substitutions that contribute to thermal stability based on comparison of the amino acid sequences of homologous proteins derived from bacteria that grow at different temperatures. A key feature of the method is that it compares the sequences based not simply on the amino acid identity, but rather on the structural and physicochemical properties of the side chain. The method accurately identified stabilizing substitutions in three well-studied systems and was validated prospectively by experimentally testing predicted stabilizing substitutions in a polyamine oxidase. In each case, the method outperformed the widely used bioinformatic consensus approach. The method can also provide insight into fundamental aspects of protein structure, for example, by identifying how many sequence positions in a given protein are relevant to temperature adaptation.

Scaffold proteins help mediate interactions between protein partners, often to optimize intracellular signaling. Herein, we use comparative, biochemical, biophysical, molecular, and cellular approaches to investigate how the scaffold protein NEMO contributes to signaling in the NF-κB pathway. Comparison of NEMO and the related protein optineurin from a variety of evolutionarily distant organisms revealed that a central region of NEMO, called the Intervening Domain (IVD), is conserved between NEMO and optineurin. Previous studies have shown that this central core region of the IVD is required for cytokine-induced activation of IκB kinase (IKK). We show that the analogous region of optineurin can functionally replace the core region of the NEMO IVD. We also show that an intact IVD is required for the formation of disulfide-bonded dimers of NEMO. Moreover, inactivating mutations in this core region abrogate the ability of NEMO to form ubiquitin-induced liquid-liquid phase separation droplets in vitro and signal-induced puncta in vivo. Thermal and chemical denaturation studies of truncated NEMO variants indicate that the IVD, while not intrinsically destabilizing, can reduce the stability of surrounding regions of NEMO, due to the conflicting structural demands imparted on this region by flanking upstream and downstream domains. This conformational strain in the IVD mediates allosteric communication between N- and C-terminal regions of NEMO. Overall, these results support a model in which the IVD of NEMO participates in signal-induced activation of the IKK/NF-κB pathway by acting as a mediator of conformational changes in NEMO.