Comparing and Validating Machine Learning Models for Mycobacterium tuberculosis Drug Discovery.

Tuberculosis is a global health dilemma. In 2016, the WHO reported 10.4 million incidences and 1.7 million deaths. The need to develop new treatments for those infected with Mycobacterium tuberculosis ( Mtb) has led to many large-scale phenotypic screens and many thousands of new active compounds identified in vitro. However, with limited funding, efforts to discover new active molecules against Mtb needs to be more efficient. Several computational machine learning approaches have been shown to have good enrichment and hit rates. We have curated small molecule Mtb data and developed new models with a total of 18,886 molecules with activity cutoffs of 10 µM, 1 µM, and 100 nM. These data sets were used to evaluate different machine learning methods (including deep learning) and metrics and to generate predictions for additional molecules published in 2017. One Mtb model, a combined in vitro and in vivo data Bayesian model at a 100 nM activity yielded the following metrics for 5-fold cross validation: accuracy = 0.88, precision = 0.22, recall = 0.91, specificity = 0.88, kappa = 0.31, and MCC = 0.41. We have also curated an evaluation set ( n = 153 compounds) published in 2017, and when used to test our model, it showed the comparable statistics (accuracy = 0.83, precision = 0.27, recall = 1.00, specificity = 0.81, kappa = 0.36, and MCC = 0.47). We have also compared these models with additional machine learning algorithms showing Bayesian machine learning models constructed with literature Mtb data generated by different laboratories generally were equivalent to or outperformed deep neural networks with external test sets. Finally, we have also compared our training and test sets to show they were suitably diverse and different in order to represent useful evaluation sets. Such Mtb machine learning models could help prioritize compounds for testing in vitro and in vivo.

A small diversity library of α-methyl amide analogs of sulindac for probing anticancer structure-activity relationships.

Non-steroidal anti-inflammatory drugs (NSAIDs) have a variety of potential indications that include management of pain and inflammation as well as chemoprevention and/or treatment of cancer. Furthermore, a specific form of ibuprofen, dexibuprofen or the S-(+) form, shows interesting neurological activities and has been proposed for the treatment of Alzheimer's disease. In a continuation of our work probing the anticancer activity of small sulindac libraries, we have prepared and screened a small diversity library of α-methyl substituted sulindac amides in the profen class. Several compounds of this series displayed promising activity compared with a lead sulindac analog.

Diverse amide analogs of sulindac for cancer treatment and prevention.

Sulindac is a non-steroidal anti-inflammatory drug (NSAID) that has shown significant anticancer activity. Sulindac sulfide amide (1) possessing greatly reduced COX-related inhibition relative to sulindac displayed in vivo antitumor activity that was comparable to sulindac in a human colon tumor xenograft model. Inspired by these observations, a panel of diverse sulindac amide derivatives have been synthesized and their activity probed against three cancer cell lines (prostate, colon and breast). A neutral analog, compound 79 was identified with comparable potency relative to lead 1 and activity against a panel of lymphoblastic leukemia cell lines. Several new series also show good activity relative to the parent (1), including five analogs that also possess nanomolar inhibitory potencies against acute lymphoblastic leukemia cells. Several new analogs identified may serve as anticancer lead candidates for further development.

Synthesis and preliminary assessment of the anticancer and Wnt/ß-catenin inhibitory activity of small amide libraries of fenamates and profens.

As part of an ongoing program to study the anticancer activity of non-steroidal anti-inflammatory drugs (NSAIDs) through generating diversity libraries of multiple NSAID scaffolds, we synthesized a series of NSAID amide derivatives and screened these sets against three cancer cell lines (prostate, colon and breast) and Wnt/ß-catenin signaling. The evaluated amide analog libraries show significant anticancer activity/cell proliferation inhibition, and specific members of the sets show inhibition of Wnt/ß-catenin signaling.

Machine Learning Model Analysis and Data Visualization with Small Molecules Tested in a Mouse Model of Mycobacterium tuberculosis Infection (2014-2015).

The renewed urgency to develop new treatments for Mycobacterium tuberculosis (Mtb) infection has resulted in large-scale phenotypic screening and thousands of new active compounds in vitro. The next challenge is to identify candidates to pursue in a mouse in vivo efficacy model as a step to predicting clinical efficacy. We previously analyzed over 70 years of this mouse in vivo efficacy data, which we used to generate and validate machine learning models. Curation of 60 additional small molecules with in vivo data published in 2014 and 2015 was undertaken to further test these models. This represents a much larger test set than for the previous models. Several computational approaches have now been applied to analyze these molecules and compare their molecular properties beyond those attempted previously. Our previous machine learning models have been updated, and a novel aspect has been added in the form of mouse liver microsomal half-life (MLM t1/2) and in vitro-based Mtb models incorporating cytotoxicity data that were used to predict in vivo activity for comparison. Our best Mtb in vivo models possess fivefold ROC values > 0.7, sensitivity > 80%, and concordance > 60%, while the best specificity value is >40%. Use of an MLM t1/2 Bayesian model affords comparable results for scoring the 60 compounds tested. Combining MLM stability and in vitro Mtb models in a novel consensus workflow in the best cases has a positive predicted value (hit rate) > 77%. Our results indicate that Bayesian models constructed with literature in vivo Mtb data generated by different laboratories in various mouse models can have predictive value and may be used alongside MLM t1/2 and in vitro-based Mtb models to assist in selecting antitubercular compounds with desirable in vivo efficacy. We demonstrate for the first time that consensus models of any kind can be used to predict in vivo activity for Mtb. In addition, we describe a new clustering method for data visualization and apply this to the in vivo training and test data, ultimately making the method accessible in a mobile app.

Open Source Bayesian Models. 1. Application to ADME/Tox and Drug Discovery Datasets.

On the order of hundreds of absorption, distribution, metabolism, excretion, and toxicity (ADME/Tox) models have been described in the literature in the past decade which are more often than not inaccessible to anyone but their authors. Public accessibility is also an issue with computational models for bioactivity, and the ability to share such models still remains a major challenge limiting drug discovery. We describe the creation of a reference implementation of a Bayesian model-building software module, which we have released as an open source component that is now included in the Chemistry Development Kit (CDK) project, as well as implemented in the CDD Vault and in several mobile apps. We use this implementation to build an array of Bayesian models for ADME/Tox, in vitro and in vivo bioactivity, and other physicochemical properties. We show that these models possess cross-validation receiver operator curve values comparable to those generated previously in prior publications using alternative tools. We have now described how the implementation of Bayesian models with FCFP6 descriptors generated in the CDD Vault enables the rapid production of robust machine learning models from public data or the user's own datasets. The current study sets the stage for generating models in proprietary software (such as CDD) and exporting these models in a format that could be run in open source software using CDK components. This work also demonstrates that we can enable biocomputation across distributed private or public datasets to enhance drug discovery.

Combining computational methods for hit to lead optimization in Mycobacterium tuberculosis drug discovery.

PURPOSE: Tuberculosis treatments need to be shorter and overcome drug resistance. Our previous large scale phenotypic high-throughput screening against Mycobacterium tuberculosis (Mtb) has identified 737 active compounds and thousands that are inactive. We have used this data for building computational models as an approach to minimize the number of compounds tested. METHODS: A cheminformatics clustering approach followed by Bayesian machine learning models (based on publicly available Mtb screening data) was used to illustrate that application of these models for screening set selections can enrich the hit rate. RESULTS: In order to explore chemical diversity around active cluster scaffolds of the dose-response hits obtained from our previous Mtb screens a set of 1924 commercially available molecules have been selected and evaluated for antitubercular activity and cytotoxicity using Vero, THP-1 and HepG2 cell lines with 4.3%, 4.2% and 2.7% hit rates, respectively. We demonstrate that models incorporating antitubercular and cytotoxicity data in Vero cells can significantly enrich the selection of non-toxic actives compared to random selection. Across all cell lines, the Molecular Libraries Small Molecule Repository (MLSMR) and cytotoxicity model identified ~10% of the hits in the top 1% screened (>10 fold enrichment). We also showed that seven out of nine Mtb active compounds from different academic published studies and eight out of eleven Mtb active compounds from a pharmaceutical screen (GSK) would have been identified by these Bayesian models. CONCLUSION: Combining clustering and Bayesian models represents a useful strategy for compound prioritization and hit-to lead optimization of antitubercular agents.

Are bigger data sets better for machine learning? Fusing single-point and dual-event dose response data for Mycobacterium tuberculosis.

Tuberculosis is a major, neglected disease for which the quest to find new treatments continues. There is an abundance of data from large phenotypic screens in the public domain against Mycobacterium tuberculosis (Mtb). Since machine learning methods can learn from past data, we were interested in addressing whether more data builds better models. We now describe using Bayesian machine learning to assess whether we can improve our models by combining the large quantities of single-point data with the much smaller (higher quality) dual-event data sets, which use both dose-response data for both whole-cell antitubercular activity and Vero cell cytotoxicity. We have evaluated 12 models ranging from different single-point, dual-event dose-response, single-point and dual-event dose-response as well as combined data sets for three distinct data sets from the same laboratory. We used a fourth data set of active and inactive compounds from the same group as well as a smaller set of 177 active compounds from GlaxoSmithKline as test sets. Our data suggest combining single-point with dual-event dose-response data does not diminish the internal or external predictive ability of the models based on the receiver operator curve (ROC) for these models (internal ROC range 0.83-0.91, external ROC range 0.62-0.83) compared to the orders of magnitude smaller dual-event models (internal ROC range 0.6-0.83 and external ROC 0.54-0.83). In conclusion, models developed with 1200-5000 compounds appear to be as predictive as those generated with 25 000-350 000 molecules. Our results have implications for justifying further high-throughput screening versus focused testing based on model predictions.

Looking back to the future: predicting in vivo efficacy of small molecules versus Mycobacterium tuberculosis.

Selecting and translating in vitro leads for a disease into molecules with in vivo activity in an animal model of the disease is a challenge that takes considerable time and money. As an example, recent years have seen whole-cell phenotypic screens of millions of compounds yielding over 1500 inhibitors of Mycobacterium tuberculosis (Mtb). These must be prioritized for testing in the mouse in vivo assay for Mtb infection, a validated model utilized to select compounds for further testing. We demonstrate learning from in vivo active and inactive compounds using machine learning classification models (Bayesian, support vector machines, and recursive partitioning) consisting of 773 compounds. The Bayesian model predicted 8 out of 11 additional in vivo actives not included in the model as an external test set. Curation of 70 years of Mtb data can therefore provide statistically robust computational models to focus resources on in vivo active small molecule antituberculars. This highlights a cost-effective predictor for in vivo testing elsewhere in other diseases.

Targeted degrader technologies as prospective SARS-CoV-2 therapies.

COVID-19 remains a severe public health threat despite the WHO declaring an end to the public health emergency in May 2023. Continual development of SARS-CoV-2 variants with resistance to vaccine-induced or natural immunity necessitates constant vigilance as well as new vaccines and therapeutics. Targeted protein degradation (TPD) remains relatively untapped in antiviral drug discovery and holds the promise of attenuating viral resistance development. From a unique structural design perspective, this review covers antiviral degrader merits and challenges by highlighting key coronavirus protein targets and their co-crystal structures, specifically illustrating how TPD strategies can refine existing SARS-CoV-2 3CL protease inhibitors to potentially produce superior protease-degrading agents.

Preparation of von Hippel-Lindau (VHL) E3 ubiquitin ligase ligands exploiting constitutive hydroxyproline for benzylic amine protection.

The von Hippel-Lindau (VHL) protein serves as the substrate recognition subunit of the multi-subunit Cullin-2 RING E3 ubiquitin ligase (CRL2VHL), which regulates intracellular concentrations of hypoxia inducible factors (HIFs) through a ubiquitin proteasome system (UPS) cascade. Strategic recruitment of CRL2VHL by bi- or trifunctional targeted protein degraders (e.g., PROTACs®) offers the prospect of promoting aberrant polyubiquitination and ensuing proteasomal degradation of disease-related proteins. Non-peptidic, l-hydroxyproline-bearing VHL ligands such as VH032 (1) and its chiral benzylic amine analog Me-VH032 (2), are functional components of targeted protein degraders commonly employed for this purpose. Herein, we compare two approaches for the preparation of 1 and 2 primarily highlighting performance differences between Pd(OAc)2 and Pd-PEPPSI-IPr for the key C-H arylation of 4-methylthiazole. Results from this comparison prompted the development of a unified, five-step route for the preparation of either VH032 (1) or Me-VH032 (2) in multigram quantities, resulting in yields of 56% and 61% for 1 and 2, respectively. Application of N-Boc-l-4-hydroxyproline rather than N-tert-butoxycarbonyl to shield the benzylic amine during the coupling step enhances step economy. Additionally, we identified previously undisclosed minor byproducts generated during arylation steps along with observations from amine deprotection and amidation reaction steps that may prove helpful not only for the preparation of 1 and 2, but for other VHL recruiting ligands, as well.

Fusing dual-event data sets for Mycobacterium tuberculosis machine learning models and their evaluation.

The search for new tuberculosis treatments continues as we need to find molecules that can act more quickly, be accommodated in multidrug regimens, and overcome ever increasing levels of drug resistance. Multiple large scale phenotypic high-throughput screens against Mycobacterium tuberculosis (Mtb) have generated dose response data, enabling the generation of machine learning models. These models also incorporated cytotoxicity data and were recently validated with a large external data set. A cheminformatics data-fusion approach followed by Bayesian machine learning, Support Vector Machine, or Recursive Partitioning model development (based on publicly available Mtb screening data) was used to compare individual data sets and subsequent combined models. A set of 1924 commercially available molecules with promising antitubercular activity (and lack of relative cytotoxicity to Vero cells) were used to evaluate the predictive nature of the models. We demonstrate that combining three data sets incorporating antitubercular and cytotoxicity data in Vero cells from our previous screens results in external validation receiver operator curve (ROC) of 0.83 (Bayesian or RP Forest). Models that do not have the highest 5-fold cross-validation ROC scores can outperform other models in a test set dependent manner. We demonstrate with predictions for a recently published set of Mtb leads from GlaxoSmithKline that no single machine learning model may be enough to identify compounds of interest. Data set fusion represents a further useful strategy for machine learning construction as illustrated with Mtb. Coverage of chemistry and Mtb target spaces may also be limiting factors for the whole-cell screening data generated to date.

6-Oxo and 6-thio purine analogs as antimycobacterial agents.

6-Oxo and 6-thio analogs of purine were prepared based on the initial activity screening of a small, diverse purine library against Mycobacterium tuberculosis (Mtb). Certain 6-oxo and 6-thio-substituted purine analogs described herein showed moderate to good inhibitory activity. N(9)-substitution apparently enhances the anti-mycobacterial activity in the purine series described herein. Several 2-amino and 2-chloro purine analogs were also synthesized that showed moderate inhibitory activity against Mtb.

Novel Non-Cyclooxygenase Inhibitory Derivative of Sulindac Inhibits Breast Cancer Cell Growth In Vitro and Reduces Mammary Tumorigenesis in Rats.

The nonsteroidal anti-inflammatory drug (NSAID) sulindac demonstrates attractive anticancer activity, but the toxicity resulting from cyclooxygenase (COX) inhibition and the suppression of physiologically important prostaglandins precludes its long-term, high dose use in the clinic for cancer prevention or treatment. While inflammation is a known tumorigenic driver, evidence suggests that sulindac's antineoplastic activity is partially or fully independent of its COX inhibitory activity. One COX-independent target proposed for sulindac is cyclic guanosine monophosphate phosphodiesterase (cGMP PDE) isozymes. Sulindac metabolites, i.e., sulfide and sulfone, inhibit cGMP PDE enzymatic activity at concentrations comparable with those associated with cancer cell growth inhibitory activity. Additionally, the cGMP PDE isozymes PDE5 and PDE10 are overexpressed during the early stages of carcinogenesis and appear essential for cancer cell proliferation and survival based on gene silencing experiments. Here, we describe a novel amide derivative of sulindac, sulindac sulfide amide (SSA), which was rationally designed to eliminate COX-inhibitory activity while enhancing cGMP PDE inhibitory activity. SSA was 68-fold and 10-fold less potent than sulindac sulfide (SS) in inhibiting COX-1 and COX-2, respectively, but 10-fold more potent in inhibiting growth and inducing apoptosis in breast cancer cells. The pro-apoptotic activity of SSA was associated with inhibition of cGMP PDE activity, elevation of intracellular cGMP levels, and activation of cGMP-dependent protein kinase (PKG) signaling, as well as the inhibition of ß-catenin/Tcf transcriptional activity. SSA displayed promising in vivo anticancer activity, resulting in a 57% reduction in the incidence and a 62% reduction in the multiplicity of tumors in the N-methyl-N-nitrosourea (MNU)-induced model of breast carcinogenesis. These findings provide strong evidence for cGMP/PKG signaling as a target for breast cancer prevention or treatment and the COX-independent anticancer properties of sulindac. Furthermore, this study validates the approach of optimizing off-target effects by reducing the COX-inhibitory activity of sulindac for future targeted drug discovery efforts to enhance both safety and efficacy.

Novel substituted pyrimidines as HCV replication (replicase) inhibitors.

Compound 1 was identified as a HCV replication inhibitor from screening/early SAR triage. Potency improvement was achieved via modulation of substituent on the 5-azo linkage. Due to potential toxicological concern, the 5-azo linkage was replaced with 5-alkenyl or 5-alkynyl moiety. Analogs containing the 5-alkynyl linkage were found to be potent inhibitors of HCV replication. Further evaluation identified compounds 53 and 63 with good overall profile, in terms of replicon potency, selectivity and in vivo characteristics. Initial target engagement studies suggest that these novel carbanucleoside-like derivatives may inhibit the HCV replication complex (replicase).

Pyridofuran substituted pyrimidine derivatives as HCV replication (replicase) inhibitors.

Introduction of nitrogen atom into the benzene ring of a previously identified HCV replication (replicase) benzofuran inhibitor 2, resulted in the discovery of the more potent pyridofuran analogue 5. Subsequent introduction of small alkyl and alkoxy ligands into the pyridine ring resulted in further improvements in replicon potency. Replacement of the 4-chloro moiety on the pyrimidine core with a methyl group, and concomitant monoalkylation of the C-2 amino moiety resulted in the identification of several inhibitors with desirable characteristics. Inhibitor 41, from the monosubstituted pyridofuran and inhibitor 50 from the disubstituted series displayed excellent potency, selectivity (GAPDH/MTS CC(50)) and PK parameters in all species studied, while the selectivity in the thymidine incorporation assay (DNA·CC(50)) was low.

Synthesis and SAR of geminal substitutions at the C5' carbosugar position of pyrimidine-derived HCV inhibitors.

The installation of geminal substitution at the C5' position of the carbosugar in our pyrimidine-derived hepatitis C inhibitor series is reported. SAR studies around the C5' position led to the installation of the dimethyl group as the optimal functionality. An improved route was subsequently designed to access these substitutions. Expanded SAR at the C2 amino position led to the utilization of C2 ethers. These compounds exhibited good potency, high selectivity, and excellent plasma exposure and bioavailability in rodent as well as in higher species.

5-Benzothiazole substituted pyrimidine derivatives as HCV replication (replicase) inhibitors.

Based on a previously identified HCV replication (replicase) inhibitor 1, SAR efforts were conducted around the pyrimidine core to improve the potency and pharmacokinetic profile of the inhibitors. A benzothiazole moiety was found to be the optimal substituent at the pyrimidine 5-position. Due to potential reactivity concern, the 4-chloro residue was replaced by a methyl group with some loss in potency and enhanced rat in vivo profile. Extensive investigations at the C-2 position resulted in identification of compound 16 that demonstrated very good replicon potency, selectivity and rodent plasma/target organ concentration. Inhibitor 16 also demonstrated good plasma levels and oral bioavailability in dogs, while monkey exposure was rather low. Chemistry optimization towards a practical route to install the benzothiazole moiety resulted in an efficient direct C-H arylation protocol.

Synthesis and SAR of pyridothiazole substituted pyrimidine derived HCV replication inhibitors.

Introduction of a nitrogen atom into the benzene ring of a previously identified HCV replication (replicase) benzothiazole inhibitor 1, resulted in the discovery of the more potent pyridothiazole analogues 3. The potency and PK properties of the compounds were attenuated by the introductions of various functionalities at the R(1), R(2) or R(3) positions of the molecule (compound 3). Inhibitors 38 and 44 displayed excellent potency, selectivity (GAPDH/MTS CC(50)), PK parameters in all species studied, and cross genotype activity.