Traumatic injury causes selective degeneration and TDP-43 mislocalization in human iPSC-derived C9orf72-associated ALS/FTD motor neurons.

A hexanucleotide repeat expansion (HRE) in C9orf72 is the most common genetic cause of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). However, patients with the HRE exhibit a wide disparity in clinical presentation and age of symptom onset suggesting an interplay between genetic background and environmental stressors. Neurotrauma as a result of traumatic brain or spinal cord injury has been shown to increase the risk of ALS/FTD in epidemiological studies. Here, we combine patient-specific induced pluripotent stem cells (iPSCs) with a custom-built device to deliver biofidelic stretch trauma to C9orf72 patient and isogenic control motor neurons (MNs) in vitro. We find that mutant but not control MNs exhibit selective degeneration after a single incident of severe trauma, which can be partially rescued by pretreatment with a C9orf72 antisense oligonucleotide. A single incident of mild trauma does not cause degeneration but leads to cytoplasmic accumulation of TDP-43 in C9orf72 MNs. This mislocalization, which only occurs briefly in isogenic controls, is eventually restored in C9orf72 MNs after 6 days. Lastly, repeated mild trauma ablates the ability of patient MNs to recover. These findings highlight alterations in TDP-43 dynamics in C9orf72 ALS/FTD patient MNs following traumatic injury and demonstrate that neurotrauma compounds neuropathology in C9orf72 ALS/FTD. More broadly, our work establishes an in vitro platform that can be used to interrogate the mechanistic interactions between ALS/FTD and neurotrauma.

Enhancing the Small-Scale Screenable Biological Space beyond Known Chemogenomics Libraries with Gray Chemical Matter─Compounds with Novel Mechanisms from High-Throughput Screening Profiles.

Phenotypic assays have become an established approach to drug discovery. Greater disease relevance is often achieved through cellular models with increased complexity and more detailed readouts, such as gene expression or advanced imaging. However, the intricate nature and cost of these assays impose limitations on their screening capacity, often restricting screens to well-characterized small compound sets such as chemogenomics libraries. Here, we outline a cheminformatics approach to identify a small set of compounds with likely novel mechanisms of action (MoAs), expanding the MoA search space for throughput limited phenotypic assays. Our approach is based on mining existing large-scale, phenotypic high-throughput screening (HTS) data. It enables the identification of chemotypes that exhibit selectivity across multiple cell-based assays, which are characterized by persistent and broad structure activity relationships (SAR). We validate the effectiveness of our approach in broad cellular profiling assays (Cell Painting, DRUG-seq, and Promotor Signature Profiling) and chemical proteomics experiments. These experiments revealed that the compounds behave similarly to known chemogenetic libraries, but with a notable bias toward novel protein targets. To foster collaboration and advance research in this area, we have curated a public set of such compounds based on the PubChem BioAssay dataset and made it available for use by the scientific community.

Compound Activity Prediction with Dose-Dependent Transcriptomic Profiles and Deep Learning.

Predicting compound activity in assays is a long-standing challenge in drug discovery. Computational models based on compound-induced gene expression signatures from a single profiling assay have shown promise toward predicting compound activity in other, seemingly unrelated, assays. Applications of such models include predicting mechanisms-of-action (MoA) for phenotypic hits, identifying off-target activities, and identifying polypharmacologies. Here, we introduce transcriptomics-to-activity transformer (TAT) models that leverage gene expression profiles observed over compound treatment at multiple concentrations to predict the compound activity in other biochemical or cellular assays. We built TAT models based on gene expression data from a RASL-seq assay to predict the activity of 2692 compounds in 262 dose-response assays. We obtained useful models for 51% of the assays, as determined through a realistic held-out set. Prospectively, we experimentally validated the activity predictions of a TAT model in a malaria inhibition assay. With a 63% hit rate, TAT successfully identified several submicromolar malaria inhibitors. Our results thus demonstrate the potential of transcriptomic responses over compound concentration and the TAT modeling framework as a cost-efficient way to identify the bioactivities of promising compounds across many assays.

Polyurethane Culture Substrates Enable Long-Term Neuron Monoculture in a Human in vitro Model of Neurotrauma.

Human induced pluripotent stem cell (hiPSC)-derived cells can reproduce human-specific pathophysiology, patient-specific vulnerability, and gene-environment interactions in neurological disease. Human in vitro models of neurotrauma therefore have great potential to advance the field. However, this potential cannot be realized until important biomaterials challenges are addressed. Status quo stretch injury models of neurotrauma culture cells on sheets of polydimethylsiloxane (PDMS) that are incompatible with long-term monoculture of hiPSC-derived neurons. Here, we overcame this challenge in an established human in vitro neurotrauma model by replacing PDMS with a highly biocompatible form of polyurethane (PU). This substitution allowed long-term monoculture of hiPSC-derived neurons. It also changed the biomechanics of stretch injury. We quantified these changes experimentally using high-speed videography and digital image correlation. We used finite element modeling to quantify the influence of the culture substrate's thickness, stiffness, and coefficient of friction on membrane stretch and concluded that the coefficient of friction explained most of the observed biomechanical changes. Despite these changes, we demonstrated that the modified model produced a robust, dose-dependent trauma phenotype in hiPSC-derived neuron monocultures. In summary, the introduction of this PU film makes it possible to maintain hiPSC-derived neurons in monoculture for long periods in a human in vitro neurotrauma model. In doing so, it opens new horizons in the field of neurotrauma by enabling the unique experimental paradigms (e.g., isogenic models) associated with hiPSC-derived neurons.

Abyssal hydrothermal springs-Cryptic incubators for brooding octopus.

Does warmth from hydrothermal springs play a vital role in the biology and ecology of abyssal animals? Deep off central California, thousands of octopus (Muusoctopus robustus) migrate through cold dark waters to hydrothermal springs near an extinct volcano to mate, nest, and die, forming the largest known aggregation of octopus on Earth. Warmth from the springs plays a key role by raising metabolic rates, speeding embryonic development, and presumably increasing reproductive success; we show that brood times for females are ~1.8 years, far faster than expected for abyssal octopods. Using a high-resolution subsea mapping system, we created landscape-scale maps and image mosaics that reveal 6000 octopus in a 2.5-ha area. Because octopuses die after reproducing, hydrothermal springs indirectly provide a food supplement to the local energy budget. Although localized deep-sea heat sources may be essential to octopuses and other warm-tolerant species, most of these unique and often cryptic habitats remain undiscovered and unexplored.

On the Choice of Active Site Sequences for Kinase-Ligand Affinity Prediction.

Recent work showed that active site rather than full-protein-sequence information improves predictive performance in kinase-ligand binding affinity prediction. To refine the notion of an "active site", we here propose and compare multiple definitions. We report significant evidence that our novel definition is superior to previous definitions and better models of ATP-noncompetitive inhibitors. Moreover, we leverage the discontiguity of the active site sequence to motivate novel protein-sequence augmentation strategies and find that combining them further improves performance.

Removal of trichloroethene from thin clay lenses by electrical resistance heating: Laboratory experiments and the effects of gas saturation.

The removal of dissolved volatile organic compounds (VOCs) from low-permeability lenses is important to limit back diffusion at sites impacted by dense non-aqueous phase liquids (DNAPLs). In situ thermal treatment (ISTT) technologies have the potential to treat DNAPL-impacted sites by enhancing diffusion from low-permeability lenses during heating. A series of two-dimensional laboratory tank experiments was conducted to investigate heating, gas formation, and trichloroethene (TCE) removal from a clay lens surrounded by sand. Results showed preferential heating of the clay and substantial TCE removal, with post-heating relative concentrations less than 0.06. The extent of TCE removal was not explained by only an increase in the aqueous TCE diffusion coefficient with increased temperature. Modelling estimates based on 1D diffusion from the lens showed that diffusion through both gas and water phases was required to match observations. Gas formation in the interior of the lens was also indicated by measured changes in bulk electrical conductivity of the clay during cool down, with gas saturations estimated to be greater than 0.21 at the end of heating. These estimates were larger than those needed to match the observed removal by diffusion, and suggest that connected gas pathways were created in the lens during heating, but that not all of the gas produced was part of those pathways. These results suggest that ISTT technologies may be effective in removing dissolved VOCs from thin clay lenses, and that gas formation within the clay should be considered when predicting the extent and rate of removal.

Collaborative Profile-QSAR: A Natural Platform for Building Collaborative Models among Competing Companies.

Massively multitask bioactivity models that transfer learning between thousands of assays have been shown to work dramatically better than separate models trained on each individual assay. In particular, the applicability domain for a given model can expand from compounds similar to those tested in that specific assay to those tested across the full complement of contributing assays. If many large companies would share their assay data and train models on the superset, predictions should be better than what each company can do alone. However, a company's compounds, targets, and activities are among their most guarded trade secrets. Strategies have been proposed to share just the individual collaborators' models, without exposing any of the training data. Profile-QSAR (pQSAR) is a two-level, multitask, stacked model. It uses profiles of level-1 predictions from single-task models for thousands of assays as compound descriptors for level-2 models. This work describes its simple and natural adaptation to safe collaboration by model sharing. Broad model sharing has not yet been implemented across multiple large companies, so there are numerous unanswered questions. Novartis was formed from several mergers and acquisitions. In principle, this should allow an internal simulation of model sharing. In practice, the lack of metadata about the origins of compounds and assays made this difficult. Nevertheless, we have attempted to simulate this process and propose some findings: multitask pQSAR is always an improvement over single-task models; collaborative multitask modeling did not improve predictions on internal compounds; collaboration did improve predictions for external compounds but far less than the purely internal multitask modeling for internal compounds; collaborative models for external compounds increasingly improve as overlap between compound collections increases; combining profiles from inside and outside the company is not best, with internal predictions better using only the inside profile and external using only the outside profile, but a consensus of models using all three profiles is best on external compounds and a good compromise on internal compounds. We anticipate similar results from other model-sharing approaches. Indeed, since collaborative pQSAR through model sharing is mathematically identical to pQSAR using actual shared data, we believe our conclusions should apply to collaborative modeling by any current method even including the unlikely scenario of directly sharing all chemical structures and assay data.

MAIP: a web service for predicting blood-stage malaria inhibitors.

Malaria is a disease affecting hundreds of millions of people across the world, mainly in developing countries and especially in sub-Saharan Africa. It is the cause of hundreds of thousands of deaths each year and there is an ever-present need to identify and develop effective new therapies to tackle the disease and overcome increasing drug resistance. Here, we extend a previous study in which a number of partners collaborated to develop a consensus in silico model that can be used to identify novel molecules that may have antimalarial properties. The performance of machine learning methods generally improves with the number of data points available for training. One practical challenge in building large training sets is that the data are often proprietary and cannot be straightforwardly integrated. Here, this was addressed by sharing QSAR models, each built on a private data set. We describe the development of an open-source software platform for creating such models, a comprehensive evaluation of methods to create a single consensus model and a web platform called MAIP available at https://www.ebi.ac.uk/chembl/maip/ . MAIP is freely available for the wider community to make large-scale predictions of potential malaria inhibiting compounds. This project also highlights some of the practical challenges in reproducing published computational methods and the opportunities that open-source software can offer to the community.

TETs compete with DNMT3 activity in pluripotent cells at thousands of methylated somatic enhancers.

Mammalian cells stably maintain high levels of DNA methylation despite expressing both positive (DNMT3A/B) and negative (TET1-3) regulators. Here, we analyzed the independent and combined effects of these regulators on the DNA methylation landscape using a panel of knockout human embryonic stem cell (ESC) lines. The greatest impact on global methylation levels was observed in DNMT3-deficient cells, including reproducible focal demethylation at thousands of normally methylated loci. Demethylation depends on TET expression and occurs only when both DNMT3s are absent. Dynamic loci are enriched for hydroxymethylcytosine and overlap with subsets of putative somatic enhancers that are methylated in ESCs and can be activated upon differentiation. We observe similar dynamics in mouse ESCs that were less frequent in epiblast stem cells (EpiSCs) and scarce in somatic tissues, suggesting a conserved pluripotency-linked mechanism. Taken together, our data reveal tightly regulated competition between DNMT3s and TETs at thousands of somatic regulatory sequences within pluripotent cells.

Biased Diversity for Effective Virtual Screening.

Virtual screening is no longer merely a matter of identifying the subset of compounds from a large collection likely to be active against a particular endpoint. This viewpoint shares some distinctive practices at Novartis, where virtual screening combines multiple computational tools that marry the competing goals of biasing the selection of compounds toward multiple desired properties, while diversifying the selection to sample the available chemistry space, identifying quality compounds that inform drug discovery. Topics include the various considerations needed for a successful virtual screening practice: triaging, compound quality, accuracy and test sets, activity prediction including multitask modeling, virtual profiling, automation, multiproperty bias, diversity and property spaces, and biased-diversity designs.

All-Assay-Max2 pQSAR: Activity Predictions as Accurate as Four-Concentration IC50s for 8558 Novartis Assays.

Profile-quantitative structure-activity relationship (pQSAR) is a massively multitask, two-step machine learning method with unprecedented scope, accuracy, and applicability domain. In step one, a "profile" of conventional single-assay random forest regression models are trained on a very large number of biochemical and cellular pIC50 assays using Morgan 2 substructural fingerprints as compound descriptors. In step two, a panel of partial least squares (PLS) models are built using the profile of pIC50 predictions from those random forest regression models as compound descriptors (hence the name). Previously described for a panel of 728 biochemical and cellular kinase assays, we have now built an enormous pQSAR from 11 805 diverse Novartis (NVS) IC50 and EC50 assays. This large number of assays, and hence of compound descriptors for PLS, dictated reducing the profile by only including random forest regression models whose predictions correlate with the assay being modeled. The random forest regression and pQSAR models were evaluated with our "realistically novel" held-out test set, whose median average similarity to the nearest training set member across the 11 805 assays was only 0.34, comparable to the novelty of compounds actually selected from virtual screens. For the 11 805 single-assay random forest regression models, the median correlation of prediction with the experiment was only rext2 = 0.05, virtually random, and only 8% of the models achieved our standard success threshold of rext2 = 0.30. For pQSAR, the median correlation was rext2 = 0.53, comparable to four-concentration experimental IC50s, and 72% of the models met our rext2 > 0.30 standard, totaling 8558 successful models. The successful models included assays from all of the 51 annotated target subclasses, as well as 4196 phenotypic assays, indicating that pQSAR can be applied to virtually any disease area. Every month, all models are updated to include new measurements, and predictions are made for 5.5 million NVS compounds, totaling 50 billion predictions. Common uses have included virtual screening, selectivity design, toxicity and promiscuity prediction, mechanism-of-action prediction, and others. Several such actual applications are described.

Dissecting the Functional Consequences of De Novo DNA Methylation Dynamics in Human Motor Neuron Differentiation and Physiology.

The somatic DNA methylation (DNAme) landscape is established early in development but remains highly dynamic within focal regions that overlap with gene regulatory elements. The significance of these dynamic changes, particularly in the central nervous system, remains unresolved. Here, we utilize a powerful human embryonic stem cell differentiation model for the generation of motor neurons (MNs) in combination with genetic mutations in the de novo DNAme machinery. We quantitatively dissect the role of DNAme in directing somatic cell fate with high-resolution genome-wide bisulfite-, bulk-, and single-cell-RNA sequencing. We find defects in neuralization and MN differentiation in DNMT3A knockouts (KO) that can be rescued by the targeting of DNAme to key developmental loci using catalytically inactive dCas9. We also find decreased dendritic arborization and altered electrophysiological properties in DNMT3A KO MNs. Our work provides a list of DNMT3A-regulated targets and a mechanistic link between de novo DNAme, cellular differentiation, and human MN function.

Physiologically Based Pharmacokinetic Modeling in Lead Optimization. 2. Rational Bioavailability Design by Global Sensitivity Analysis To Identify Properties Affecting Bioavailability.

When medicinal chemists need to improve oral bioavailability (%F) during lead optimization, they systematically modify compound properties mainly based on their own experience and general rules of thumb. However, at least a dozen properties can influence %F, and the difficulty of multiparameter optimization for such complex nonlinear processes grows combinatorially with the number of variables. Furthermore, strategies can be in conflict. For example, adding a polar or charged group will generally increase solubility but decrease permeability. Identifying the 2 or 3 properties that most influence %F for a given compound series would make %F optimization much more efficient. We previously reported an adaptation of physiologically based pharmacokinetic (PBPK) simulations to predict %F for lead series from purely computational inputs within a 2-fold average error. Here, we run thousands of such simulations to generate a comprehensive "bioavailability landscape" for each series. A key innovation was recognition that the large and variable number of p Ka's in drug molecules could be replaced by just the two straddling the isoelectric point. Another was use of the ZINC database to cull out chemically inaccessible regions of property space. A quadratic partial least squares regression (PLS) accurately fits a continuous surface to these thousands of bioavailability predictions. The PLS coefficients indicate the globally sensitive compound properties. The PLS surface also displays the %F landscape in these sensitive properties locally around compounds of particular interest. Finally, being quick to calculate, the PLS equation can be combined with models for activity and other properties for multiobjective lead optimization.

Physiologically Based Pharmacokinetic Modeling in Lead Optimization. 1. Evaluation and Adaptation of GastroPlus To Predict Bioavailability of Medchem Series.

When medicinal chemists need to improve bioavailability (%F) within a chemical series during lead optimization, they synthesize new series members with systematically modified properties mainly by following experience and general rules of thumb. More quantitative models that predict %F of proposed compounds from chemical structure alone have proven elusive. Global empirical %F quantitative structure-property (QSPR) models perform poorly, and projects have too little data to train local %F QSPR models. Mechanistic oral absorption and physiologically based pharmacokinetic (PBPK) models simulate the dissolution, absorption, systemic distribution, and clearance of a drug in preclinical species and humans. Attempts to build global PBPK models based purely on calculated inputs have not achieved the <2-fold average error needed to guide lead optimization. In this work, local GastroPlus PBPK models are instead customized for individual medchem series. The key innovation was building a local QSPR for a numerically fitted effective intrinsic clearance (CLloc). All inputs are subsequently computed from structure alone, so the models can be applied in advance of synthesis. Training CLloc on the first 15-18 rat %F measurements gave adequate predictions, with clear improvements up to about 30 measurements, and incremental improvements beyond that.

Profile-QSAR 2.0: Kinase Virtual Screening Accuracy Comparable to Four-Concentration IC50s for Realistically Novel Compounds.

While conventional random forest regression (RFR) virtual screening models appear to have excellent accuracy on random held-out test sets, they prove lacking in actual practice. Analysis of 18 historical virtual screens showed that random test sets are far more similar to their training sets than are the compounds project teams actually order. A new, cluster-based "realistic" training/test set split, which mirrors the chemical novelty of real-life virtual screens, recapitulates the poor predictive power of RFR models in real projects. The original Profile-QSAR (pQSAR) method greatly broadened the domain of applicability over conventional models by using as independent variables a profile of activity predictions from all historical assays in a large protein family. However, the accuracy still fell short of experiment on realistic test sets. The improved "pQSAR 2.0" method replaces probabilities of activity from naïve Bayes categorical models at several thresholds with predicted IC50s from RFR models. Unexpectedly, the high accuracy also requires removing the RFR model for the actual assay of interest from the independent variable profile. With these improvements, pQSAR 2.0 activity predictions are now statistically comparable to medium-throughput four-concentration IC50 measurements even on the realistic test set. Beyond the yes/no activity predictions from a typical high-throughput screen (HTS) or conventional virtual screen, these semiquantitative IC50 predictions allow for predicted potency, ligand efficiency, lipophilic efficiency, and selectivity against antitargets, greatly facilitating hitlist triaging and enabling virtual screening panels such as toxicity panels and overall promiscuity predictions.

Blockade of orexin receptors attenuates the cardiovascular response to air-jet stress in spontaneously hypertensive rats.

This study tested the hypothesis that orexin plays a role in the elevated pressor response to acute stress in the spontaneously hypertensive rat (SHR). The pressor response to air jet stress (AJS) (n=11/group) was 2.5 times greater in vehicle treated SHR versus Wistar (WIS) rats. Systemic delivery of 30mg/kg of the dual orexin receptor antagonist almorexant did not significantly change resting mean arterial pressure (MAP) but did attenuate the pressor response elicited by AJS to a greater extent in the SHR compared to the Wistar rats (~65% versus ~33% reduction respectively; n=6/group). Alternatively 100mg/kg almorexant reduced resting MAP in the SHR (~25mm Hg drop) and attenuated both the heart rate (HR; ~50% reduction) and MAP (~62% reduction) response to AJS in both strains (n=6/group). Systemic application of SB-334867 (3mg/kg), an orexin receptor type 1 antagonist (n=5/group), selectively reduced resting MAP and attenuated the HR response to AJS in the SHR but had no effect on the pressor response in either strain. The potential role of endogenous orexin release in cardiovascular control in the SHR was linked to a significant increase in brain-derived neurotrophic factor mRNA expression in the hypothalamus and elevated orexin receptor expression (type 2 only) in the dorsal pons when compared to WIS (n=4/group). These results demonstrate that the exaggerated pressor response in the SHR to stress is linked to increased orexin receptor activation and possibly altered orexin receptor expression in the dorsal pons and BDNF expression in the hypothalamus.

Chronic heart failure alters orexin and melanin concentrating hormone but not corticotrophin releasing hormone-related gene expression in the brain of male Lewis rats.

OBJECTIVE: The aim of this study was to investigate the effect of chronic heart failure (HF; 16 weeks post left coronary artery ligation) on the brain's orexin (ORX) and related neuropeptide systems. METHODS: Indicators of cardiac function, including the percent fractional shortening (%FS) left ventricular posterior wall shortening velocity (LVPWSV) were assessed via echocardiography at 16 weeks post myocardial infarction or sham treatment in male Lewis rats (n=5/group). Changes in gene expression in HF versus control (CON) groups were quantified by real-time PCR in the hypothalamus, amygdala and dorsal pons. RESULTS: HF significantly reduced both the %FS and LVPWSV when compared to CON animals (P<0.02). In the hypothalamus ORX gene expression was significantly reduced in HF and correlated with changes in cardiac function when compared to CON (P<0.02). No significant changes in hypothalamic ORX receptor (type 1 or type 2) gene expression were identified. Alternatively hypothalamic melanin concentrating hormone (MCH) gene expression was significantly upregulated in HF animals and negatively correlated with LVPWSV (P<0.006). In both the amygdala and dorsal pons ORX type 2 receptor expression was significantly down-regulated in HF compared to CON. ORX receptor type 1, CRH and CRH type 1 and type 2 receptor expressions were unchanged by HF in all brain regions analyzed. CONCLUSION: These observations support previous work demonstrating that cardiovascular disease modulates the ORX system and identify that in the case of chronic HF the ORX system is altered in parallel with changes in MCH expression but independent of any significant changes in the central CRH system. This raises the new possibility that ORX and MCH systems may play an important role in the pathophysiology of HF.

Surrogate AutoShim: predocking into a universal ensemble kinase receptor for three dimensional activity prediction, very quickly, without a crystal structure.

"Ensemble surrogate AutoShim" is a kinase specific extension of the AutoShim docking method that solves the three traditional limitations of conventional docking: (1) it gives good correlations with affinity, (2) does not require a target protein structure, and (3) for a preprocessed company archive of 1.5 million compounds, is as fast as traditional 2D QSAR. It does require several hundred experimental IC 50 values for each new target. Original AutoShim adds pharmacophore "shims" to a crystal structure binding site. An iterative partial least squares (PLS) procedure selects the best pose, while adjusting the shim weights to reproduce IC 50 data. Surrogate AutoShim adjusts shims in one crystal structure to reproduce IC 50 data for a different kinase target. Ensemble surrogate AutoShim uses 16 structurally diverse kinase crystal structures as a "universal ensemble kinase receptor", suitable for any kinase target. The 1.5 million member Novartis screening collection has been predocked into the shimmed ensemble, so new kinase models can be built, and the entire corporate archive virtually screened, in hours rather than weeks. A kinase-biased set of 10,000 compounds, that samples the entire corporate archive, has been designed for lead discovery by iterative kinase screening.

AutoShim: empirically corrected scoring functions for quantitative docking with a crystal structure and IC50 training data.

It has been notoriously difficult to develop general all-purpose scoring functions for high-throughput docking that correlate with measured binding affinity. As a practical alternative, AutoShim uses the program Magnet to add point-pharmacophore like "shims" to the binding site of each protein target. The pharmacophore shims are weighted by partial least-squares (PLS) regression, adjusting the all-purpose scoring function to reproduce IC 50 data, much as the shims in an NMR magnet are weighted to optimize the field for a better spectrum. This dramatically improves the affinity predictions on 25% of the compounds held out at random. An iterative procedure chooses the best pose during the process of shim parametrization. This method reproducibly converges to a consistent solution, regardless of starting pose, in just 2-4 iterations, so these robust models do not overtrain. Sets of complex multifeature shims, generated by a recursive partitioning method, give the best activity predictions, but these are difficult to interpret when designing new compounds. Sets of simpler single-point pharmacophores still predict affinity reasonably well and clearly indicate the molecular interactions producing effective binding. The pharmacophore requirements are very reproducible, irrespective of the compound sets used for parametrization, lending confidence to the predictions and interpretations. The automated procedure does require a training set of experimental compounds but otherwise adds little extra time over conventional docking.