Lesion-remote astrocytes govern microglia-mediated white matter repair.

Spared regions of the damaged central nervous system undergo dynamic remodeling and exhibit a remarkable potential for therapeutic exploitation. Here, lesion-remote astrocytes (LRAs), which interact with viable neurons, glia and neural circuitry, undergo reactive transformations whose molecular and functional properties are poorly understood. Using multiple transcriptional profiling methods, we interrogated LRAs from spared regions of mouse spinal cord following traumatic spinal cord injury (SCI). We show that LRAs acquire a spectrum of molecularly distinct, neuroanatomically restricted reactivity states that evolve after SCI. We identify transcriptionally unique reactive LRAs in degenerating white matter that direct the specification and function of local microglia that clear lipid-rich myelin debris to promote tissue repair. Fueling this LRA functional adaptation is Ccn1 , which encodes for a secreted matricellular protein. Loss of astrocyte CCN1 leads to excessive, aberrant activation of local microglia with (i) abnormal molecular specification, (ii) dysfunctional myelin debris processing, and (iii) impaired lipid metabolism, culminating in blunted debris clearance and attenuated neurological recovery from SCI. Ccn1 -expressing white matter astrocytes are specifically induced by local myelin damage and generated in diverse demyelinating disorders in mouse and human, pointing to their fundamental, evolutionarily conserved role in white matter repair. Our findings show that LRAs assume regionally divergent reactivity states with functional adaptations that are induced by local context-specific triggers and influence disorder outcome. Astrocytes tile the central nervous system (CNS) where they serve vital roles that uphold healthy nervous system function, including regulation of synapse development, buffering of neurotransmitters and ions, and provision of metabolic substrates 1 . In response to diverse CNS insults, astrocytes exhibit disorder-context specific transformations that are collectively referred to as reactivity 2-5 . The characteristics of regionally and molecularly distinct reactivity states are incompletely understood. The mechanisms through which distinct reactivity states arise, how they evolve or resolve over time, and their consequences for local cell function and CNS disorder progression remain enigmatic. Immediately adjacent to CNS lesions, border-forming astrocytes (BFAs) undergo transcriptional reprogramming and proliferation to form a neuroprotective barrier that restricts inflammation and supports axon regeneration 6-9 . Beyond the lesion, spared but dynamic regions of the injured CNS exhibit varying degrees of synaptic circuit remodeling and progressive cellular responses to secondary damage that have profound consequences for neural repair and recovery 10,11 . Throughout these cytoarchitecturally intact, but injury-reactive regions, lesion-remote astrocytes (LRAs) intermingle with neurons and glia, undergo little to no proliferation, and exhibit varying degrees of cellular hypertrophy 7,12,13 . The molecular and functional properties of LRAs remain grossly undefined. Therapeutically harnessing spared regions of the injured CNS will require a clearer understanding of the accompanying cellular and molecular landscape. Here, we leveraged integrative transcriptional profiling methodologies to identify multiple spatiotemporally resolved, molecularly distinct states of LRA reactivity within the injured spinal cord. Computational modeling of LRA-mediated heterotypic cell interactions, astrocyte-specific conditional gene deletion, and multiple mouse models of acute and chronic CNS white matter degeneration were used to interrogate a newly identified white matter degeneration-reactive astrocyte subtype. We define how this reactivity state is induced and its role in governing the molecular and functional specification of local microglia that clear myelin debris from the degenerating white matter to promote repair.

Comprehensive Lipidomic Automation Workflow using Large Language Models.

Lipidomics generates large data that makes manual annotation and interpretation challenging. Lipid chemical and structural diversity with structural isomers further complicates annotation. Although, several commercial and open-source software for targeted lipid identification exists, it lacks automated method generation workflows and integration with statistical and bioinformatics tools. We have developed the Comprehensive Lipidomic Automated Workflow (CLAW) platform with integrated workflow for parsing, detailed statistical analysis and lipid annotations based on custom multiple reaction monitoring (MRM) precursor and product ion pair transitions. CLAW contains several modules including identification of carbon-carbon double bond position(s) in unsaturated lipids when combined with ozone electrospray ionization (OzESI)-MRM methodology. To demonstrate the utility of the automated workflow in CLAW, large-scale lipidomics data was collected with traditional and OzESI-MRM profiling on biological and non-biological samples. Specifically, a total of 1497 transitions organized into 10 MRM-based mass spectrometry methods were used to profile lipid droplets isolated from different brain regions of 18-24 month-old Alzheimer's disease mice and age-matched wild-type controls. Additionally, triacyclglycerols (TGs) profiles with carbon-carbon double bond specificity were generated from canola oil samples using OzESI-MRM profiling. We also developed an integrated language user interface with large language models using artificially intelligent (AI) agents that permits users to interact with the CLAW platform using a chatbot terminal to perform statistical and bioinformatic analyses. We envision CLAW pipeline to be used in high-throughput lipid structural identification tasks aiding users to generate automated lipidomics workflows ranging from data acquisition to AI agent-based bioinformatic analysis.

Modeling Molecular Mechanisms of Pirfenidone Interaction with Kinases.

Scar formation is a process that occurs due to increased collagen deposition and uncontrolled inflammation. Previous studies have demonstrated that Pirfenidone (Pf), an FDA approved anti-inflammatory and antifibrotic drug can reduce inflammation in vivo as well as regulate activation of LPS-stimulated neutrophils. However, the molecular level mechanism of Pf's action is not well understood. Here, we used neural networks to identify new targets and molecular modeling methods to investigate the Pf's action pathways at the molecular level that are related to its ability to reduce both the inflammatory and remodeling phases of the wound healing process. Out of all the potential targets identified, both molecular docking and molecular dynamics results suggest that Pf has a noteworthy binding preference towards the active conformation of the p38 mitogen activated protein kinase-14 (MAPK14) and it is potentially a type I inhibitor-like molecule. In addition to p38 MAPK (MAPK14), additional potential targets of Pf include AKT1, MAP3K4, MAP2K3, MAP2K6, MSK2, MAP2K2, ERK1, ERK2, and PDK1. We conclude that several proteins/kinases, rather than a single target, are involved in Pf's wound healing ability to regulate signaling, inflammation, and proliferation.

Automated Bioanalytical Workflow for Ligand Binding-Based Pharmacokinetic Assay Development.

The growth of therapeutic monoclonal antibodies (mAbs) continues to accelerate due to their success as treatments for many diseases. As new therapeutics are developed, it is increasingly important to have robust bioanalytical methods to measure the pharmacokinetics (PK) of circulating therapeutic mAbs in serum. Ligand-binding assays such as enzyme-linked immunosorbent assays (ELISAs) with anti-idiotypic antibodies (anti-IDs) targeting the variable regions of the therapeutic antibody are sensitive and specific bioanalytical methods to measure levels of therapeutic antibodies in a biological matrix. However, soluble circulating drug mAb targets can interfere with the anti-IDs binding to the therapeutic mAb, thereby resulting in an underestimation of total drug concentration. Therefore, in addition to a high binding affinity for the mAb, the selection of anti-IDs and the assay format that are not impacted by soluble antigens and have low matrix interference is essential for developing a robust PK assay. Standardized automated approaches to screen and select optimal reagents and assay formats are critical to increase efficiency, quality, and PK assay robustness. However, there does not exist an integrated screening and analysis platform to develop robust PK assays across multiple formats. We have developed an automated workflow and scoring platform with multiple bioanalytical assay parameters that allow for ranking of candidate anti-IDs. A primary automated indirect electrochemiluminescence (ECL) was utilized to shortlist the anti-IDs that were selected for labeling and screening in pairs. A secondary screen using an ECL sandwich assay with labeled-anti-ID pairings was used to test multiple PK assay formats to identify the best anti-ID pairing/PK assay format. We developed an automated assay using fixed plate maps combined with a human-guided graphical user interface-based scoring system and compared it to a data-dependent scoring system using Gaussian mixture models for automated scoring and selection. Our approach allowed for screening of anti-IDs and identification of the most robust PK assay format with significantly reduced time and resources compared with traditional approaches. We believe that such standardized, automated, and integrated platforms that accelerate the development of PK assays will become increasingly important for supporting future human clinical trials.

Amyloid ß Induces Lipid Droplet-Mediated Microglial Dysfunction in Alzheimer's Disease.

Several microglia-expressed genes have emerged as top risk variants for Alzheimer's disease (AD). Impaired microglial phagocytosis is one of the main proposed outcomes by which these AD-risk genes may contribute to neurodegeneration, but the mechanisms translating genetic association to cellular dysfunction remain unknown. Here we show that microglia form lipid droplets (LDs) upon exposure to amyloid-beta (Aß), and that their LD load increases with proximity to amyloid plaques in brains from human patients and the AD mouse model 5xFAD. LD formation is dependent upon age and disease progression and is more prominent in the hippocampus in mice and humans. Despite variability in LD load between microglia from male versus female animals and between cells from different brain regions, LD-laden microglia exhibited a deficit in Aß phagocytosis. Unbiased lipidomic analysis identified a substantial decrease in free fatty acids (FFAs) and a parallel increase in triacylglycerols (TAGs) as the key metabolic transition underlying LD formation. We demonstrate that DGAT2, a key enzyme for the conversion of FFAs to TAGs, promotes microglial LD formation, is increased in microglia from 5xFAD and human AD brains, and that inhibiting DGAT2 improved microglial uptake of Aß. These findings identify a new lipid-mediated mechanism underlying microglial dysfunction that could become a novel therapeutic target for AD.

Variability in phenylalanine side chain conformations facilitates broad substrate tolerance of fatty acid binding in cockroach milk proteins.

Diploptera punctata, also known as the Pacific beetle cockroach, is a viviparous cockroach that gives birth to live offspring and secretes a highly concentrated mixture of glycosylated proteins as a source of nourishment for developing embryos. These proteins are lipocalins that bind to lipids and crystallize in the gut of the embryo. A structure of milk crystals harvested from the embryos showed that the milk-derived crystals were heterogeneous and made of three proteins (called Lili-Mips). We hypothesized that the isoforms of Lili-Mip would display different affinities for fatty acids due to the ability of the pocket to bind multiple acyl chain lengths. We previously reported the structures of Lili-Mip from crystals grown in vivo and recombinantly expressed Lili-Mip2. These structures are similar, and both bind to several fatty acids. This study explores the specificity and affinity of fatty acid binding to recombinantly expressed Lili-Mip 1, 2 & 3. We show that all isoforms can bind to different fatty acids with similar affinities. We also report the thermostability of Lili-Mip is pH dependent, where stability is highest at acidic pH and declines as the pH increases to physiological levels near 7.0. We show that thermostability is an inherent property of the protein, and glycosylation and ligand binding do not change it significantly. Measuring the pH in the embryo's gut lumen and gut cells suggests that the pH in the gut is acidic and the pH inside the gut cells is closer to neutral pH. In various crystal structures (reported here and previously by us), Phe-98 and Phe-100 occupy multiple conformations in the binding pocket. In our earlier work, we had shown that the loops at the entrance could adapt various conformations to change the size of the binding pocket. Here we show Phe-98 and Phe-100 can reorient to stabilize interactions at the bottom of the cavity-and change the volume of the cavity from 510 Å3 to 337 Å3. Together they facilitate the binding of fatty acids of different acyl chain lengths.

Learning Relationships Between Chemical and Physical Stability for Peptide Drug Development.

PURPOSE OR OBJECTIVE: Chemical and physical stabilities are two key features considered in pharmaceutical development. Chemical stability is typically reported as a combination of potency and degradation product. Moreover, fluorescent reporter Thioflavin-T is commonly used to measure physical stability. Executing stability studies is a lengthy process and requires extensive resources. To reduce the resources and shorten the process for stability studies during the development of a drug product, we introduce a machine learning-based model for predicting the chemical stability over time using both formulation conditions as well as aggregation curves. METHODS: In this work, we develop the relationships between the formulation, stability timepoint, and the chemical stability measurements and evaluated the performance on a random test set. We have developed a multilayer perceptron (MLP) for total degradation prediction and a random forest (RF) model for potency. RESULTS: The coefficient of determination (R2) of 0.945 and a mean absolute error (MAE) of 0.421 were achieved on the test set when using MLP for total degradation. Similarly, we achieved a R2 of 0.908 and MAE of 1.435 when predicting potency using the RF model. When physical stability measurements are included into the MLP model, the MAE of predicting TD decreases to 0.148. Using a similar strategy for potency prediction, the MAE decreases to 0.705 for the RF model. CONCLUSIONS: We conclude two important points: first, chemical stability can be modeled using machine learning techniques and second there is a relationship between the physical stability of a peptide and its chemical stability.

Identification of Monomethyl Branched-Chain Lipids by a Combination of Liquid Chromatography Tandem Mass Spectrometry and Charge-Switching Chemistries.

While various mass spectrometric approaches have been applied to lipid analysis, unraveling the extensive structural diversity of lipids remains a significant challenge. Notably, these approaches often fail to differentiate between isomeric lipids─a challenge that is particularly acute for branched-chain fatty acids (FAs) that often share similar (or identical) mass spectra to their straight-chain isomers. Here, we utilize charge-switching strategies that combine ligated magnesium dications with deprotonated fatty acid anions. Subsequent activation of these charge inverted anions yields mass spectra that differentiate anteiso-branched- from straight-chain and iso-branched-chain FA isomers with the predictable fragmentation enabling de novo assignment of anteiso branch points. The application of these charge-inversion chemistries in both gas- and solution-phase modalities is demonstrated to assign the position of anteiso-methyl branch-points in FAs and, with the aid of liquid chromatography, can be extended to de novo assignment of additional branching sites via predictable fragmentation patterns as methyl branching site(s) move closer to the carboxyl carbon. The gas-phase approach is shown to be compatible with top-down structure elucidation of complex lipids such as phosphatidylcholines, while the integration of solution-phase charge-inversion with reversed phase liquid chromatography enables separation and unambiguous identification of FA structures within isomeric mixtures. Taken together, the presented charge-switching MS-based technique, in combination with liquid chromatography, enables the structural identification of branched-chain FA without the requirement of authentic methyl-branched FA reference standards.

One Scaffold, Different Organelle Sensors: pH-Activable Fluorescent Probes for Targeting Live Microglial Cell Organelles.

Targeting live cell organelles is essential for imaging, understanding, and controlling specific biochemical processes. Typically, fluorescent probes with distinct structural scaffolds are used to target specific cell organelles. Here, we have designed a modular one-step synthetic strategy using a common reaction intermediate to develop new lysosomal, mitochondrial, and nucleus-targeting pH-activable fluorescent probes that are all based on a single boron dipyrromethane scaffold. The divergent cell organelle targeting was achieved by synthesizing probes with specific functional group changes to the central scaffold resulting in differential fluorescence and pKa . Specifically, we show that the functional group transformation of the same scaffold influences cellular localization and specificity of pH-activable fluorescent probes in live primary microglial cells with pKa values ranging from ∼3.2-6.0. We introduce a structure-organelle-relationship (SOR) framework to target nuclei (NucShine), lysosomes (LysoShine), and mitochondria (MitoShine) in live microglia. This work will result in future applications of SOR beyond imaging to target and control organelle-specific biochemical processes in disease-specific models.

Neurotoxic reactive astrocytes induce cell death via saturated lipids.

Astrocytes regulate the response of the central nervous system to disease and injury and have been hypothesized to actively kill neurons in neurodegenerative disease1-6. Here we report an approach to isolate one component of the long-sought astrocyte-derived toxic factor5,6. Notably, instead of a protein, saturated lipids contained in APOE and APOJ lipoparticles mediate astrocyte-induced toxicity. Eliminating the formation of long-chain saturated lipids by astrocyte-specific knockout of the saturated lipid synthesis enzyme ELOVL1 mitigates astrocyte-mediated toxicity in vitro as well as in a model of acute axonal injury in vivo. These results suggest a mechanism by which astrocytes kill cells in the central nervous system.

Monitoring phagocytic uptake of amyloid ß into glial cell lysosomes in real time.

Phagocytosis by glial cells is essential to regulate brain function during health and disease. Therapies for Alzheimer's disease (AD) have primarily focused on targeting antibodies to amyloid ß (Aß) or inhibitng enzymes that make it, and while removal of Aß by phagocytosis is protective early in AD it remains poorly understood. Impaired phagocytic function of glial cells during later stages of AD likely contributes to worsened disease outcome, but the underlying mechanisms of how this occurs remain unknown. We have developed a human Aß1-42 analogue (AßpH) that exhibits green fluorescence upon internalization into the acidic organelles of cells but is non-fluorescent at physiological pH. This allowed us to image, for the first time, glial uptake of AßpH in real time in live animals. We find that microglia phagocytose more AßpH than astrocytes in culture, in brain slices and in vivo. AßpH can be used to investigate the phagocytic mechanisms responsible for removing Aß from the extracellular space, and thus could become a useful tool to study Aß clearance at different stages of AD.

Accurate Prediction of Inhibitor Binding to HIV-1 Protease Using CANDOCK.

The human immunodeficiency virus 1 (HIV-1) protease is an important target for treating HIV infection. Our goal was to benchmark a novel molecular docking protocol and determine its effectiveness as a therapeutic repurposing tool by predicting inhibitor potency to this target. To accomplish this, we predicted the relative binding scores of various inhibitors of the protease using CANDOCK, a hierarchical fragment-based docking protocol with a knowledge-based scoring function. We first used a set of 30 HIV-1 protease complexes as an initial benchmark to optimize the parameters for CANDOCK. We then compared the results from CANDOCK to two other popular molecular docking protocols Autodock Vina and Smina. Our results showed that CANDOCK is superior to both of these protocols in terms of correlating predicted binding scores to experimental binding affinities with a Pearson coefficient of 0.62 compared to 0.48 and 0.49 for Vina and Smina, respectively. We further leveraged the Database of Useful Decoys: Enhanced (DUD-E) HIV protease set to ascertain the effectiveness of each protocol in discriminating active versus decoy ligands for proteases. CANDOCK again displayed better efficacy over the other commonly used molecular docking protocols with area under the receiver operating characteristic curve (AUROC) of 0.94 compared to 0.71 and 0.74 for Vina and Smina. These findings support the utility of CANDOCK to help discover novel therapeutics that effectively inhibit HIV-1 and possibly other retroviral proteases.

Accelerated Reactivity Mechanism and Interpretable Machine Learning Model of N-Sulfonylimines toward Fast Multicomponent Reactions.

We introduce chemical reactivity flowcharts to help chemists interpret reaction outcomes using statistically robust machine learning models trained on a small number of reactions. We developed fast N-sulfonylimine multicomponent reactions for understanding reactivity and to generate training data. Accelerated reactivity mechanisms were investigated using density functional theory. Intuitive chemical features learned by the model accurately predicted heterogeneous reactivity of N-sulfonylimine with different carboxylic acids. Validation of the predictions shows that reaction outcome interpretation is useful for human chemists.

Targeting polyamine biosynthesis to stimulate beta cell regeneration in zebrafish.

Type 1 diabetes (T1D) is a disease characterized by destruction of the insulin-producing beta cells. Currently, there remains a critical gap in our understanding of how to reverse or prevent beta cell loss in individuals with T1D. Previous studies in mice discovered that pharmacologically inhibiting polyamine biosynthesis using difluoromethylornithine (DFMO) resulted in preserved beta cell function and mass. Similarly, treatment of non-obese diabetic mice with the tyrosine kinase inhibitor Imatinib mesylate reversed diabetes. The promising findings from these animal studies resulted in the initiation of two separate clinical trials that would repurpose either DFMO (NCT02384889) or Imatinib (NCT01781975) and determine effects on diabetes outcomes; however, whether these drugs directly stimulated beta cell growth remained unknown. To address this, we used the zebrafish model system to determine pharmacological impact on beta cell regeneration. After induction of beta cell death, zebrafish embryos were treated with either DFMO or Imatinib. Neither drug altered whole-body growth or exocrine pancreas length. Embryos treated with Imatinib showed no effect on beta cell regeneration; however, excitingly, DFMO enhanced beta cell regeneration. These data suggest that pharmacological inhibition of polyamine biosynthesis may be a promising therapeutic option to stimulate beta cell regeneration in the setting of diabetes.

DUBS: A Framework for Developing Directory of Useful Benchmarking Sets for Virtual Screening.

Benchmarking is a crucial step in evaluating virtual screening methods for drug discovery. One major issue that arises among benchmarking data sets is a lack of a standardized format for representing the protein and ligand structures used to benchmark the virtual screening method. To address this, we introduce the Directory of Useful Benchmarking Sets (DUBS) framework, as a simple and flexible tool to rapidly create benchmarking sets using the protein databank. DUBS uses a simple input text based format along with the Lemon data mining framework to efficiently access and organize data to the protein databank and output commonly used inputs for virtual screening software. The simple input format used by DUBS allows users to define their own benchmarking data sets and access the corresponding information directly from the software package. Currently, it only takes DUBS less than 2 min to create a benchmark using this format. Since DUBS uses a simple python script, users can easily modify this to create more complex benchmarks. We hope that DUBS will be a useful community resource to provide a standardized representation for benchmarking data sets in virtual screening. The DUBS package is available on GitHub at https://github.com/chopralab/lemon/tree/master/dubs.

cando.py: Open Source Software for Predictive Bioanalytics of Large Scale Drug-Protein-Disease Data.

Traditional drug discovery methods focus on optimizing the efficacy of a drug against a single biological target of interest for a specific disease. However, evidence supports the multitarget theory, i.e., drugs work by exerting their therapeutic effects via interaction with multiple biological targets, which have multiple phenotypic effects. Analytics of drug-protein interactions on a large proteomic scale provides insight into disease systems while also allowing for prediction of putative therapeutics against specific indications. We present a Python package for analysis of drug-proteome and drug-disease relationships implementing the Computational Analysis of Novel Drug Opportunities (CANDO) platform. The CANDO package allows for rapid drug similarity assessment, most notably via an in-house interaction scoring protocol where billions of drug-protein interactions are rapidly scored and the similarity of drug-proteome interaction signatures is calculated. The package also implements a variety of benchmarking protocols for shotgun drug discovery and repurposing, i.e., to determine how every known drug is related to every other in the context of the indications/diseases for which they are approved. Drug predictions are generated through consensus scoring of the most similar compounds to drugs known to treat a particular indication. Support for comparing and ranking novel chemical entities, as well as machine learning modules for both benchmarking and putative drug candidate prediction is also available. The CANDO Python package is available on GitHub at https://github.com/ram-compbio/CANDO, through the Conda Python package installer, and at http://compbio.org/software/.

A 2-Tyr-1-carboxylate Mononuclear Iron Center Forms the Active Site of a Paracoccus Dimethylformamidase.

N,N-dimethyl formamide (DMF) is an extensively used organic solvent but is also a potent pollutant. Certain bacterial species from genera such as Paracoccus, Pseudomonas, and Alcaligenes have evolved to use DMF as a sole carbon and nitrogen source for growth via degradation by a dimethylformamidase (DMFase). We show that DMFase from Paracoccus sp. strain DMF is a halophilic and thermostable enzyme comprising a multimeric complex of the α2 ß2 or (α2 ß2 )2 type. One of the three domains of the large subunit and the small subunit are hitherto undescribed protein folds of unknown evolutionary origin. The active site consists of a mononuclear iron coordinated by two Tyr side-chain phenolates and one carboxylate from Glu. The Fe3+ ion in the active site catalyzes the hydrolytic cleavage of the amide bond in DMF. Kinetic characterization reveals that the enzyme shows cooperativity between subunits, and mutagenesis and structural data provide clues to the catalytic mechanism.

CANDOCK: Chemical Atomic Network-Based Hierarchical Flexible Docking Algorithm Using Generalized Statistical Potentials.

Small-molecule docking has proven to be invaluable for drug design and discovery. However, existing docking methods have several limitations such as improper treatment of the interactions of essential components in the chemical environment of the binding pocket (e.g., cofactors, metal ions, etc.), incomplete sampling of chemically relevant ligand conformational space, and the inability to consistently correlate docking scores of the best binding pose with experimental binding affinities. We present CANDOCK, a novel docking algorithm, that utilizes a hierarchical approach to reconstruct ligands from an atomic grid using graph theory and generalized statistical potential functions to sample biologically relevant ligand conformations. Our algorithm accounts for protein flexibility, solvent, metal ions, and cofactor interactions in the binding pocket that are traditionally ignored by current methods. We evaluate the algorithm on the PDBbind, Astex, and PINC proteins to show its ability to reproduce the binding mode of the ligands that is independent of the initial ligand conformation in these benchmarks. Finally, we identify the best selector and ranker potential functions such that the statistical score of the best selected docked pose correlates with the experimental binding affinities of the ligands for any given protein target. Our results indicate that CANDOCK is a generalized flexible docking method that addresses several limitations of current docking methods by considering all interactions in the chemical environment of a binding pocket for correlating the best-docked pose with biological activity. CANDOCK along with all structures and scripts used for benchmarking is available at https://github.com/chopralab/candock_benchmark.