MolScore: a scoring, evaluation and benchmarking framework for generative models in de novo drug design.

Generative models are undergoing rapid research and application to de novo drug design. To facilitate their application and evaluation, we present MolScore. MolScore already contains many drug-design-relevant scoring functions commonly used in benchmarks such as, molecular similarity, molecular docking, predictive models, synthesizability, and more. In addition, providing performance metrics to evaluate generative model performance based on the chemistry generated. With this unification of functionality, MolScore re-implements commonly used benchmarks in the field (such as GuacaMol, MOSES, and MolOpt). Moreover, new benchmarks can be created trivially. We demonstrate this by testing a chemical language model with reinforcement learning on three new tasks of increasing complexity related to the design of 5-HT2a ligands that utilise either molecular descriptors, 266 pre-trained QSAR models, or dual molecular docking. Lastly, MolScore can be integrated into an existing Python script with just three lines of code. This framework is a step towards unifying generative model application and evaluation as applied to drug design for both practitioners and researchers. The framework can be found on GitHub and downloaded directly from the Python Package Index.Scientific ContributionMolScore is an open-source platform to facilitate generative molecular design and evaluation thereof for application in drug design. This platform takes important steps towards unifying existing benchmarks, providing a platform to share new benchmarks, and improves customisation, flexibility and usability for practitioners over existing solutions.

Boosting the Ionic Conductivity of Pyrrolidinium-Based Ionic Plastic Crystals by LLZO Fillers.

Organic ionic plastic crystals (OIPCs) have attracted attention as novel organic solid electrolyte materials, but their insufficient mechanical strength and ionic conductivity have prevented their application. In this study, a lithium salt, lithium bis(fluorosulfonyl)amide (LiFSA), and an inorganic solid electrolyte, Li7La3Zr2O12 (LLZO), were added to an OIPC, N,N-diethylpyrrolidinium bis(fluorosulfonyl)amide ([C2epyr][FSA]). The fabricated organic-inorganic hybrid solid electrolytes were evaluated thermally, mechanically, and electrochemically to reveal which factors affect the properties of the electrolytes. All samples showed excellent thermal stability regardless of LiFSA or LLZO concentration, and they were found to be highly plastic and ion-conductive solids at a wide range of temperatures. It was also revealed that the addition of LLZO raised the nanoindentation stiffness (HIT) of the [C2epyr][FSA]/LiFSA composites. The ionic conductivity of the hybrid electrolytes was higher than that of the pristine OIPC, reaching a value of 2.1 × 10-4 S cm-1 at 25 °C upon addition of appropriate amounts of LiFSA and LLZO. Overall, samples with higher LiFSA concentration and moderate LLZO concentration exhibited higher ionic conductivity. Cyclic voltammetry results showed that the [C2epyr][FSA]/LiFSA/LLZO composites were lithium-ion conductors. These findings indicate that by optimizing the concentrations of lithium salt and LLZO, it would be possible to realize their applications as solid electrolytes.

Development of an effective two-equation turbulence modeling approach for simulating aerosol deposition across a range of turbulence levels.

Pharmaceutical aerosol systems present a significant challenge to computational fluid dynamics (CFD) modeling based on the need to capture multiple levels of turbulence, frequent transition between laminar and turbulent flows, anisotropic turbulent particle dispersion, and near-wall particle transport phenomena often within geometrically complex systems over multiple time scales. Two-equation turbulence models, such as the k-ω family of approximations, offer a computationally efficient solution approach, but are known to require the use of near-wall (NW) corrections and eddy interaction model (EIM) modifications for accurate predictions of aerosol deposition. The objective of this study was to develop an efficient and effective two-equation turbulence modeling approach that enables accurate predictions of pharmaceutical aerosol deposition across a range of turbulence levels. Key systems considered were the traditional aerosol deposition benchmark cases of a 90-degree bend (Re=6,000) and a vertical straight section of pipe (Re=10,000), as well as a highly complex case of direct-to-infant (D2I) nose-to-lung pharmaceutical aerosol delivery from an air-jet dry powder inhaler (DPI) including a patient interface and infant nasal geometry through mid-trachea (500

Novel Compound Heterozygous ZAP70 R37G A507T Mutations in Infant with Severe Immunodeficiency.

Zeta-chain associated protein kinase 70 kDa (ZAP70) combined immunodeficiency (CID) is an autosomal recessive severe immunodeficiency that is characterized by abnormal T-cell receptor signaling. Children with the disorder typically present during the first year of life with diarrhea, failure to thrive, and recurrent bacterial, viral, or opportunistic infections. To date, the only potential cure is hematopoietic stem cell transplant (HSCT). The majority of described mutations causing disease occur in the homozygous state, though heterozygotes are reported without a clear understanding as to how the individual mutations interact to cause disease. This case describes an infant with novel ZAP-70 deficiency mutations involving the SH2 and kinase domains cured with allogeneic HSCT utilizing a reduced-intensity conditioning regimen and graft manipulation. We then were able to further elucidate the molecular signaling alterations imparted by these mutations that lead to altered immune function. In order to examine the effect of these novel compound ZAP70 heterozygous mutations on T cells, Jurkat CD4+ T cells were transfected with either wild type, or with individual ZAP70 R37G and A507T mutant constructs. Downstream TCR signaling events and protein localization results link these novel mutations to the expected immunological outcome as seen in the patient's primary cells. This study further characterizes mutations in the ZAP70 gene as combined immunodeficiency and the clinical phenotype.

On the difficulty of validating molecular generative models realistically: a case study on public and proprietary data.

While a multitude of deep generative models have recently emerged there exists no best practice for their practically relevant validation. On the one hand, novel de novo-generated molecules cannot be refuted by retrospective validation (so that this type of validation is biased); but on the other hand prospective validation is expensive and then often biased by the human selection process. In this case study, we frame retrospective validation as the ability to mimic human drug design, by answering the following question: Can a generative model trained on early-stage project compounds generate middle/late-stage compounds de novo? To this end, we used experimental data that contains the elapsed time of a synthetic expansion following hit identification from five public (where the time series was pre-processed to better reflect realistic synthetic expansions) and six in-house project datasets, and used REINVENT as a widely adopted RNN-based generative model. After splitting the dataset and training REINVENT on early-stage compounds, we found that rediscovery of middle/late-stage compounds was much higher in public projects (at 1.60%, 0.64%, and 0.21% of the top 100, 500, and 5000 scored generated compounds) than in in-house projects (where the values were 0.00%, 0.03%, and 0.04%, respectively). Similarly, average single nearest neighbour similarity between early- and middle/late-stage compounds in public projects was higher between active compounds than inactive compounds; however, for in-house projects the converse was true, which makes rediscovery (if so desired) more difficult. We hence show that the generative model recovers very few middle/late-stage compounds from real-world drug discovery projects, highlighting the fundamental difference between purely algorithmic design and drug discovery as a real-world process. Evaluating de novo compound design approaches appears, based on the current study, difficult or even impossible to do retrospectively.Scientific Contribution This contribution hence illustrates aspects of evaluating the performance of generative models in a real-world setting which have not been extensively described previously and which hopefully contribute to their further future development.

Protomer selectivity of type II RAF inhibitors within the RAS/RAF complex.

RAF dimer inhibitors offer therapeutic potential in RAF- and RAS-driven cancers. The utility of such drugs is predicated on their capacity to occupy both RAF protomers in the RAS-RAF signaling complex. Here we describe a method to conditionally quantify drug-target occupancy at selected RAF protomers within an active RAS-RAF complex in cells. RAF target engagement can be measured in the presence or absence of any mutant KRAS allele, enabling the high-affinity state of RAF dimer inhibitors to be quantified in the cellular milieu. The intracellular protomer selectivity of clinical-stage type II RAF inhibitors revealed that ARAF protomer engagement, but not engagement of BRAF or CRAF, is commensurate with inhibition of MAPK signaling in various mutant RAS cell lines. Our results support a fundamental role for ARAF in mutant RAS signaling and reveal poor ARAF protomer vulnerability for a cohort of RAF inhibitors undergoing clinical evaluation.

Ex vivo modulation of intact tumor fragments with anti-PD-1 and anti-CTLA-4 influences the expansion and specificity of tumor-infiltrating lymphocytes.

Checkpoint inhibition (CPI) therapy and adoptive cell therapy with autologous tumor-infiltrating lymphocytes (TIL-based ACT) are the two most effective immunotherapies for the treatment of metastatic melanoma. While CPI has been the dominating therapy in the past decade, TIL-based ACT is beneficial for individuals even after progression on previous immunotherapies. Given that notable differences in response have been made when used as a subsequent treatment, we investigated how the qualities of TILs changed when the ex vivo microenvironment of intact tumor fragments were modulated with checkpoint inhibitors targeting programmed death receptor 1 (PD-1) and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4). Initially, we show that unmodified TILs from CPI-resistant individuals can be produced, are overwhelmingly terminally differentiated, and are capable of responding to tumor. We then investigate these properties in ex vivo checkpoint modulated TILs finding that that they retain these qualities. Lastly, we confirmed the specificity of the TILs to the highest responding tumor antigens, and identified this reactivity resides largely in CD39+CD69+ terminally differentiated populations. Overall, we found that anti-PD-1 will alter the proliferative capacity while anti-CTLA4 will influence breadth of antigen specificity.

Transvenous lead extractions in a single high-volume center over a 24-year period: High success rate and low complication rate.

Background: Transvenous lead extraction (TLE) procedures can be complicated and are associated with a small but significant risk of cardiovascular complications. However, methods and tools vary among centers. Objective: The purpose of this study was to the present the methods and results of pacemaker and implantable cardioverter-defibrillator TLE procedures in our center over a 24-year period. Methods: From April 1997 through 2020, we attempted to extract 2964 leads in 1780 procedures and 1642 patients. We mainly utilized single sheath technique using snaring or mechanical rotational sheaths and steel sheaths when necessary. Difficult procedures were performed by an experienced cardiologist, and close supervision was emphasized. Most of the extractions were performed using local anesthesia with sedation. Results: Median age of patients was 65.0 [interquartile range 20.00] years, and median dwelling time of leads was 5.0 [7.0] years. Clinical success was achieved in 1739 procedures (97.7%) and complete technical success in 2841 leads (95.8%). Clinical success (leaving <4 cm of the lead in the body and achieving the clinical goal for the patient) was achieved for 79 leads (2.7%). TLE failed in 44 leads (1.1%) and 41 procedures (2.3%) among 36 patients (2.2%). There were 23 cases (1.3%) of major complications, with only 1 death directly related to the procedure (<0.1%). In addition, 2 patients with sepsis died within the first 24 hours after the procedure. No caval tears occurred. Conclusion: Single sheath lead extractions utilizing snaring or mechanical rotational sheaths were effective and safe in our high-volume center as performed by experienced operators.

Defective Mitochondrial Dynamics and Protein Degradation Pathways Underlie Cadmium-Induced Neurotoxicity and Cell Death in Huntington's Disease Striatal Cells.

Exposure to heavy metals, including cadmium (Cd), can induce neurotoxicity and cell death. Cd is abundant in the environment and accumulates in the striatum, the primary brain region selectively affected by Huntington's disease (HD). We have previously reported that mutant huntingtin protein (mHTT) combined with chronic Cd exposure induces oxidative stress and promotes metal dyshomeostasis, resulting in cell death in a striatal cell model of HD. To understand the effect of acute Cd exposure on mitochondrial health and protein degradation pathways, we hypothesized that expression of mHTT coupled with acute Cd exposure would cooperatively alter mitochondrial bioenergetics and protein degradation mechanisms in striatal STHdh cells to reveal novel pathways that augment Cd cytotoxicity and HD pathogenicity. We report that mHTT cells are significantly more susceptible to acute Cd-induced cell death as early as 6 h after 40 µM CdCl2 exposure compared with wild-type (WT). Confocal microscopy, biochemical assays, and immunoblotting analysis revealed that mHTT and acute Cd exposure synergistically impair mitochondrial bioenergetics by reducing mitochondrial potential and cellular ATP levels and down-regulating the essential pro-fusion proteins MFN1 and MFN2. These pathogenic effects triggered cell death. Furthermore, Cd exposure increases the expression of autophagic markers, such as p62, LC3, and ATG5, and reduces the activity of the ubiquitin-proteasome system to promote neurodegeneration in HD striatal cells. Overall, these results reveal a novel mechanism to further establish Cd as a pathogenic neuromodulator in striatal HD cells via Cd-triggered neurotoxicity and cell death mediated by an impairment in mitochondrial bioenergetics and autophagy with subsequent alteration in protein degradation pathways.

Integrating structure-based approaches in generative molecular design.

Generative molecular design for drug discovery and development has seen a recent resurgence promising to improve the efficiency of the design-make-test-analyse cycle; by computationally exploring much larger chemical spaces than traditional virtual screening techniques. However, most generative models thus far have only utilized small-molecule information to train and condition de novo molecule generators. Here, we instead focus on recent approaches that incorporate protein structure into de novo molecule optimization in an attempt to maximize the predicted on-target binding affinity of generated molecules. We summarize these structure integration principles into either distribution learning or goal-directed optimization and for each case whether the approach is protein structure-explicit or implicit with respect to the generative model. We discuss recent approaches in the context of this categorization and provide our perspective on the future direction of the field.

Health Care Access and Lived Experience of American Indian/Alaska Native Two Spirit and LGBTQ+ Participants in the Pride and Connectedness Survey, 2020.

OBJECTIVE: To better understand health experiences among Two Spirit and lesbian, gay, bisexual, transgender, and queer and questioning (LGBTQ+) American Indian/Alaska Native (AI/AN) people, we examined experiences with access to health care of 223 AI/AN Two Spirit and LGBTQ+ people. METHODS: Participants of the Pride and Connectedness 2020 survey, conducted through the Northwest Portland Area Indian Health Board, were asked about barriers to seeking and accessing care through a 10-question scale. We compared cisgender and gender-diverse participant demographic and scale responses to explore potential differences based on gender identity using the Pearson χ2 test of independence and ordinal logistic regression, respectively. RESULTS: Both cisgender and gender-diverse participants experienced at least some difficulties accessing health care. Finances, lack of psychologists/other mental health support, and lack of psychological support groups for Two Spirit and LGBTQ+ communities were the top 3 barriers to care experienced by all participants (84%, 82%, and 80%, respectively). Compared with cisgender participants, gender-diverse participants were more likely to report difficulties accessing care for nearly all questions on the 10-question scale and nearly 3 times more likely to report fear of being mistreated within the health care system based on their gender identity (adjusted odds ratio = 2.9; 95% CI, 1.8-4.9; P < .001). CONCLUSIONS: Increased access to mental health services and improved health care provider training that focuses on culturally relevant and gender-affirming practices would benefit the health and well-being of AI/AN people who identify as Two Spirit and LGBTQ+.

Scoping review into models of interconception care delivered at well-child visits for the Australian context.

BACKGROUND: The interconception period provides an opportunity to address women's health risks and optimise birth spacing before the next pregnancy. This scoping review aimed to identify models of interconception care (ICC) delivered at well-child visits (WCVs) around the world, review the impacts of ICC delivered, and what the feasibility and applicability of these models were. METHODS: The global review included clinical studies that that were identified using medical subject headings (MeSH) and keyword combinations. Studies were included if they met the criteria: were clinical studies; examined a model of ICC; were conducted by a registered health professional; and examined women who had given birth within the last 24-months. The following databases were searched: Medline (OVID); CINAHL (EBSCO); PubMed; and Embase (OVID). Relevant studies were screened in Covidence and the data was then extracted using a narrative analysis. RESULTS: Fifteen studies met the inclusion criteria. The benefits of ICC delivered at WCVs included screening for maternal health behaviours and conditions and increase women's uptake of interventions. The studies identified that implementing ICC at WCVs was acceptable to women. Identified challenges included lack of time for health providers, lack of education among women and health providers, and limited funding for WCVs. CONCLUSION: ICC interventions found in this review included family planning counselling and provision of long-acting contraception; health promotion of folic acid; and postpartum depression screening. The research concluded that ICC delivered at WCVs contributes to improving health behaviours for future pregnancies. Increased capacity for this care at WCVs could be achieved with targeted resources and time allocation.

Near Elimination of In Vitro Predicted Extrathoracic Aerosol Deposition in Children Using a Spray-Dried Antibiotic Formulation and Pediatric Air-Jet DPI.

PURPOSE: This study evaluated the in vitro aerosol performance of a dry powder antibiotic product that combined a highly dispersible tobramycin powder with a previously optimized pediatric air-jet dry powder inhaler (DPI) across a subject age range of 2-10 years. METHODS: An excipient enhanced growth (EEG) formulation of the antibiotic tobramycin (Tobi) was prepared using a small particle spray drying technique that included mannitol as the hygroscopic excipient and trileucine as the dispersion enhancer. The Tobi-EEG formulation was aerosolized using a positive-pressure pediatric air-jet DPI that included a 3D rod array. Realistic in vitro experiments were conducted in representative airway models consistent with children in the age ranges of 2-3, 5-6 and 9-10 years using oral or nose-to-lung administration, non-humidified or humidified airway conditions, and constant or age-specific air volumes. RESULTS: Across all conditions tested, mouth-throat depositional loss was < 1% and nose-throat depositional loss was < 3% of loaded dose. Lung delivery efficiency was in the range of 77.3-85.1% of loaded dose with minor variations based on subject age (~ 8% absolute difference), oral or nasal administration (< 2%), and delivered air volume (< 2%). Humidified airway conditions had an insignificant impact on extrathoracic depositional loss and significantly increased aerosol size at the exit of a representative lung chamber. CONCLUSIONS: In conclusion, the inhaled antibiotic product nearly eliminated extrathoracic depositional loss, demonstrated high efficiency nose-to-lung antibiotic aerosol delivery in pediatric airway models for the first time, and provided ~ 80% lung delivery efficiency with little variability across subject age and administered air volume.

Augmented Hill-Climb increases reinforcement learning efficiency for language-based de novo molecule generation.

A plethora of AI-based techniques now exists to conduct de novo molecule generation that can devise molecules conditioned towards a particular endpoint in the context of drug design. One popular approach is using reinforcement learning to update a recurrent neural network or language-based de novo molecule generator. However, reinforcement learning can be inefficient, sometimes requiring up to 105 molecules to be sampled to optimize more complex objectives, which poses a limitation when using computationally expensive scoring functions like docking or computer-aided synthesis planning models. In this work, we propose a reinforcement learning strategy called Augmented Hill-Climb based on a simple, hypothesis-driven hybrid between REINVENT and Hill-Climb that improves sample-efficiency by addressing the limitations of both currently used strategies. We compare its ability to optimize several docking tasks with REINVENT and benchmark this strategy against other commonly used reinforcement learning strategies including REINFORCE, REINVENT (version 1 and 2), Hill-Climb and best agent reminder. We find that optimization ability is improved ~ 1.5-fold and sample-efficiency is improved ~ 45-fold compared to REINVENT while still delivering appealing chemistry as output. Diversity filters were used, and their parameters were tuned to overcome observed failure modes that take advantage of certain diversity filter configurations. We find that Augmented Hill-Climb outperforms the other reinforcement learning strategies used on six tasks, especially in the early stages of training or for more difficult objectives. Lastly, we show improved performance not only on recurrent neural networks but also on a reinforcement learning stabilized transformer architecture. Overall, we show that Augmented Hill-Climb improves sample-efficiency for language-based de novo molecule generation conditioning via reinforcement learning, compared to the current state-of-the-art. This makes more computationally expensive scoring functions, such as docking, more accessible on a relevant timescale.

PARP inhibition is a modulator of anti-tumor immune response in BRCA-deficient tumors.

PARP inhibitors are synthetically lethal with BRCA1/2 mutations, and in this setting, accumulation of DNA damage leads to cell death. Because increased DNA damage and subsequent immune activation can prime an anti-tumor immune response, we studied the impact of olaparib ± immune checkpoint blockade (ICB) on anti-tumor activity and the immune microenvironment. Concurrent combination of olaparib, at clinically relevant exposures, with ICB gave durable and deeper anti-tumor activity in the Brca1m BR5 model vs. monotherapies. Olaparib and combination treatment modulated the immune microenvironment, including increases in CD8+ T cells and NK cells, and upregulation of immune pathways, including type I IFN and STING signaling. Olaparib also induced a dose-dependent upregulation of immune pathways, including JAK/STAT, STING and type I IFN, in the tumor cell compartment of a BRCA1m (HBCx-10) but not a BRCA WT (HBCx-9) breast PDX model. In vitro, olaparib induced BRCAm tumor cell-specific dendritic cell transactivation. Relevance to human disease was assessed using patient samples from the MEDIOLA (NCT02734004) trial, which showed increased type I IFN, STING, and JAK/STAT pathway expression following olaparib treatment, in line with preclinical findings. These data together provide evidence for a mechanism and schedule underpinning potential benefit of ICB combination with olaparib.

KRAS is vulnerable to reversible switch-II pocket engagement in cells.

Current small-molecule inhibitors of KRAS(G12C) bind irreversibly in the switch-II pocket (SII-P), exploiting the strong nucleophilicity of the acquired cysteine as well as the preponderance of the GDP-bound form of this mutant. Nevertheless, many oncogenic KRAS mutants lack these two features, and it remains unknown whether targeting the SII-P is a practical therapeutic approach for KRAS mutants beyond G12C. Here we use NMR spectroscopy and a cellular KRAS engagement assay to address this question by examining a collection of SII-P ligands from the literature and from our own laboratory. We show that the SII-Ps of many KRAS hotspot (G12, G13, Q61) mutants are accessible using noncovalent ligands, and that this accessibility is not necessarily coupled to the GDP state of KRAS. The results we describe here emphasize the SII-P as a privileged drug-binding site on KRAS and unveil new therapeutic opportunities in RAS-driven cancer.

Patient-Centered Discussion on End-of-Life Care for Patients with Advanced COPD.

Exacerbations of chronic obstructive pulmonary disease (COPD) may lead to a rapid decline in health and subsequent death, an unfortunate tyranny of having COPD-an irreversible health condition of 16 million individuals in the USA totaling 60 million in the world. While COPD is the third largest leading cause of death, causing 3.23 million deaths worldwide in 2019 (according to the WHO), most patients with COPD do not receive adequate treatment at the end stages of life. Although death is inevitable, the trajectory towards end-of-life is less predictable in severe COPD. Thus, clinician-patient discussion for end-of-life and palliative care could bring a meaningful life-prospective to patients with advanced COPD. Here, we summarized the current understanding and treatment of COPD. This review also highlights the importance of patient-centered discussion and summarizes current status of managing patients with advanced COPD.

Evaluation of the Polyhedral Mesh Style for Predicting Aerosol Deposition in Representative Models of the Conducting Airways.

A critical factor affecting the accuracy of Computational Fluid Dynamic (CFD) simulations and the time required to conduct them is construction of the computational mesh. This study aimed to evaluate the relatively new polyhedral mesh style for simulating aerosol deposition in the upper conducting airways compared with established meshing techniques and experimental data. Hexahedral and polyhedral mesh solutions were compared in two benchmark geometries: 1) a 90°-bend with flow characteristics similar to the extrathoracic airways of an adolescent child, and 2) a double bifurcation representing bifurcations B3-B5 in an adult. Both 4-block and 5-block hexahedral meshes were used in the 90°-bend to capture the potential of fully-structured hexahedral meshes. In the 90°-bend, polyhedral elements matched polydisperse in vitro deposition data with 20% relative error (RE; averaged across the particle sizes considered), which is an improvement on the accuracy of the 4-block hexahedral mesh (35% RE) and is similar to the accuracy of the 5-block hexahedral mesh (19% RE). In the double bifurcation, deposition fraction relative differences evaluated between polyhedral and hexahedral meshes ranged from 0.3% to 28.6% for the different particle sizes assessed, which is an order of magnitude improvement compared with previous studies that considered hexahedral vs. hybrid tetrahedral-prism meshes for the same flow field. Solution convergence time with polyhedral elements was found to be 50% to 140% higher than with hexahedral meshes of comparable size. While application dependent, the increase in simulation time observed with polyhedral meshes will likely be outweighed by the ease and convenience of polyhedral mesh construction. It was concluded that the polyhedral mesh style, with sufficient resolution especially near the walls, is an excellent alternative to the highly regarded hexahedral mesh style for predicting upper airway aerosol transport and deposition and provides a powerful new tool in the assessment of respiratory aerosol dosimetry.

Applications of Artificial Intelligence in Drug Design: Opportunities and Challenges.

Artificial intelligence (AI) has undergone rapid development in recent years and has been successfully applied to real-world problems such as drug design. In this chapter, we review recent applications of AI to problems in drug design including virtual screening, computer-aided synthesis planning, and de novo molecule generation, with a focus on the limitations of the application of AI therein and opportunities for improvement. Furthermore, we discuss the broader challenges imposed by AI in translating theoretical practice to real-world drug design; including quantifying prediction uncertainty and explaining model behavior.