Are we fitting data or noise? Analysing the predictive power of commonly used datasets in drug-, materials-, and molecular-discovery.

Data-driven techniques for establishing quantitative structure property relations are a pillar of modern materials and molecular discovery. Fuelled by the recent progress in deep learning methodology and the abundance of new algorithms, it is tempting to chase benchmarks and incrementally build ever more capable machine learning (ML) models. While model evaluation has made significant progress, the intrinsic limitations arising from the underlying experimental data are often overlooked. In the chemical sciences data collection is costly, thus datasets are small and experimental errors can be significant. These limitations of such datasets affect their predictive power, a fact that is rarely considered in a quantitative way. In this study, we analyse commonly used ML datasets for regression and classification from drug discovery, molecular discovery, and materials discovery. We derived maximum and realistic performance bounds for nine such datasets by introducing noise based on estimated or actual experimental errors. We then compared the estimated performance bounds to the reported performance of leading ML models in the literature. Out of the nine datasets and corresponding ML models considered, four were identified to have reached or surpassed dataset performance limitations and thus, they may potentially be fitting noise. More generally, we systematically examine how data range, the magnitude of experimental error, and the number of data points influence dataset performance bounds. Alongside this paper, we release the Python package NoiseEstimator and provide a web-based application for computing realistic performance bounds. This study and the resulting tools will help practitioners in the field understand the limitations of datasets and set realistic expectations for ML model performance. This work stands as a reference point, offering analysis and tools to guide development of future ML models in the chemical sciences.

Switching Cytolytic Nanopores into Antimicrobial Fractal Ruptures by a Single Side Chain Mutation.

Disruption of cell membranes is a fundamental host defense response found in virtually all forms of life. The molecular mechanisms vary but generally lead to energetically favored circular nanopores. Here, we report an elaborate fractal rupture pattern induced by a single side-chain mutation in ultrashort (8-11-mers) helical peptides, which otherwise form transmembrane pores. In contrast to known mechanisms, this mode of membrane disruption is restricted to the upper leaflet of the bilayer where it exhibits propagating fronts of peptide-lipid interfaces that are strikingly similar to viscous instabilities in fluid flow. The two distinct disruption modes, pores and fractal patterns, are both strongly antimicrobial, but only the fractal rupture is nonhemolytic. The results offer wide implications for elucidating differential membrane targeting phenomena defined at the nanoscale.

Accelerated antimicrobial discovery via deep generative models and molecular dynamics simulations.

The de novo design of antimicrobial therapeutics involves the exploration of a vast chemical repertoire to find compounds with broad-spectrum potency and low toxicity. Here, we report an efficient computational method for the generation of antimicrobials with desired attributes. The method leverages guidance from classifiers trained on an informative latent space of molecules modelled using a deep generative autoencoder, and screens the generated molecules using deep-learning classifiers as well as physicochemical features derived from high-throughput molecular dynamics simulations. Within 48 days, we identified, synthesized and experimentally tested 20 candidate antimicrobial peptides, of which two displayed high potency against diverse Gram-positive and Gram-negative pathogens (including multidrug-resistant Klebsiella pneumoniae) and a low propensity to induce drug resistance in Escherichia coli. Both peptides have low toxicity, as validated in vitro and in mice. We also show using live-cell confocal imaging that the bactericidal mode of action of the peptides involves the formation of membrane pores. The combination of deep learning and molecular dynamics may accelerate the discovery of potent and selective broad-spectrum antimicrobials.

Membrane Binding of Antimicrobial Peptides Is Modulated by Lipid Charge Modification.

Peptide interactions with lipid bilayers play a key role in a range of biological processes and depend on electrostatic interactions between charged amino acids and lipid headgroups. Antimicrobial peptides (AMPs) initiate the killing of bacteria by binding to and destabilizing their membranes. The multiple peptide resistance factor (MprF) provides a defense mechanism for bacteria against a broad range of AMPs. MprF reduces the negative charge of bacterial membranes through enzymatic conversion of the anionic lipid phosphatidyl glycerol (PG) to either zwitterionic alanyl-phosphatidyl glycerol (Ala-PG) or cationic lysyl-phosphatidyl glycerol (Lys-PG). The resulting change in the membrane charge is suggested to reduce the binding of AMPs to membranes, thus impeding downstream AMP activity. Using coarse-grained molecular dynamics to investigate the effects of these modified lipids on AMP binding to model membranes, we show that AMPs have substantially reduced affinity for model membranes containing Ala-PG or Lys-PG. More than 5000 simulations in total are used to define the relationship between lipid bilayer composition, peptide sequence (using five different membrane-active peptides), and peptide binding to membranes. The degree of interaction of a peptide with a membrane correlates with the membrane surface charge density. Free energy profile (potential of mean force) calculations reveal that the lipid modifications due to MprF alter the energy barrier to peptide helix penetration of the bilayer. These results will offer a guide to the design of novel peptides, which addresses the issue of resistance via MprF-mediated membrane modification.

Membrane Permeability in Cyclic Peptides is Modulated by Core Conformations.

Cyclic peptides have the potential to bind to challenging targets, which are undruggable with small molecules, but their application is limited by low membrane permeability. Here, using a series of cyclic pentapeptides, we showed that established physicochemical criteria of permeable peptides are heavily violated. We revealed that a dominant core conformation, stabilized by amides' shielding pattern, could guide the design of novel compounds. As a result, counter-intuitive strategies, such as incorporation of polar residues, can be beneficial for permeability. We further find that core globularity is a promising descriptor, which can extend the capability of standard predictive models.

Accelerating molecular discovery through data and physical sciences: Applications to peptide-membrane interactions.

Simulation and data analysis have evolved into powerful methods for discovering and understanding molecular modes of action and designing new compounds to exploit these modes. The combination provides a strong impetus to create and exploit new tools and techniques at the interfaces between physics, biology, and data science as a pathway to new scientific insight and accelerated discovery. In this context, we explore the rational design of novel antimicrobial peptides (short protein sequences exhibiting broad activity against multiple species of bacteria). We show how datasets can be harvested to reveal features which inform new design concepts. We introduce new analysis and visualization tools: a graphical representation of the k-mer spectrum as a fundamental property encoded in antimicrobial peptide databases and a data-driven representation to illustrate membrane binding and permeation of helical peptides.

Structure and hydrogen bonding at the limits of liquid water stability.

Liquid water exhibits unconventional behaviour across its wide range of stability - from its unusually high liquid-vapour critical point down to its melting point and below where it reaches a density maximum and exhibits negative thermal expansion allowing ice to float. Understanding the molecular underpinnings of these anomalies presents a challenge motivating the study of water for well over a century. Here we examine the molecular structure of liquid water across its range of stability, from mild supercooling to the negative pressure and high temperature regimes. We use a recently-developed, electronically-responsive model of water, constructed from gas-phase molecular properties and incorporating many-body, long-range interactions to all orders; as a result the model has been shown to have high transferability from ice to the supercritical regime. We report a link between the anomalous thermal expansion of water and the behaviour of its second coordination shell and an anomaly in hydrogen bonding, which persists throughout liquid water's range of stability - from the high temperature limit of liquid water to its supercooled regime.

Signature properties of water: Their molecular electronic origins.

Water challenges our fundamental understanding of emergent materials properties from a molecular perspective. It exhibits a uniquely rich phenomenology including dramatic variations in behavior over the wide temperature range of the liquid into water's crystalline phases and amorphous states. We show that many-body responses arising from water's electronic structure are essential mechanisms harnessed by the molecule to encode for the distinguishing features of its condensed states. We treat the complete set of these many-body responses nonperturbatively within a coarse-grained electronic structure derived exclusively from single-molecule properties. Such a "strong coupling" approach generates interaction terms of all symmetries to all orders, thereby enabling unique transferability to diverse local environments such as those encountered along the coexistence curve. The symmetries of local motifs that can potentially emerge are not known a priori. Consequently, electronic responses unfiltered by artificial truncation are then required to embody the terms that tip the balance to the correct set of structures. Therefore, our fully responsive molecular model produces, a simple, accurate, and intuitive picture of water's complexity and its molecular origin, predicting water's signature physical properties from ice, through liquid-vapor coexistence, to the critical point.

Hydrogen bonding and molecular orientation at the liquid-vapour interface of water.

We determine the molecular structure and orientation at the liquid-vapour interface of water using an electronically coarse grained model constructed to include all long-range electronic responses within Gaussian statistics. The model, fit to the properties of the isolated monomer and dimer, is sufficiently responsive to generate the temperature dependence of the surface tension from ambient conditions to the critical point. Acceptor hydrogen bonds are shown to be preferentially truncated at the free surface under ambient conditions and a related asymmetry in hydrogen bonding preference is identified in bulk water. We speculate that this bonding asymmetry in bulk water is the microscopic origin of the observed surface structure.