BenevolentAI (United Kingdom)

De novo molecular design is effective for optimizing compounds with desirable properties in drug discovery. In the realm of chemoinformatics, population-based evolutionary approaches have been known for over a decade including those that use retrosynthetically inspired fragment-based workflows to to balance the generation of synthesizable molecules with a sufficiently expressive molecular generator. We describe FragGT, a fragment-based de novo molecular design approach that uses an evolutionary algorithm to optimize molecules towards an objective function. We rely on the concept of gene types to hop between related molecular fragments where a fragment's gene type is defined by the nature of its connections to adjacent groups. We show that this simple approach achieves competitive performance on the GuacaMol benchmark suite with a low computational cost and the additional benefits of transparency and practicality.

Chemical diversity is challenging to describe objectively. Despite this, various notions of chemical diversity are used throughout the medicinal chemistry optimization process in drug discovery. In this work, we show the usefulness of considering exploited vectors during different phases of the drug design process to provide a quantitative and objective description of chemical diversity. We have developed a concise and fast approach to enumerate and analyze the exploited vector patterns (EVPs) of molecular compound series, which can then be used in archetypal compound selection tasks from hit matter identification to hit expansion and lead optimization. We firstly show that EVPs can be used to assess the progressibility of compounds in a fragment library design exercise. By considering EVPs, we then show how a set of compounds can be prioritized for hit expansion using EVP-based, customizable diversity sampling approaches, reducing the time taken and mitigating human biases. We also show that EVPs are a useful tool to analyze SAR data, offering the chance to uncover correlations between different vectors without pre-determining the molecular scaffold structures. The codes used to perform these tasks are presented as easy-to-use Jupyter notebooks, which can be readily adapted for further related tasks.

ABSTRACTOver the last few years, many machine learning-based scoring functions for predicting the binding of small molecules to proteins have been developed. Their objective is to approximate the distribution which takes two molecules as input and outputs the energy of their interaction. Only a scoring function that accounts for the interatomic interactions involved in binding can accurately predict binding affinity on unseen molecules. However, many scoring functions make predictions based on dataset biases rather than an understanding of the physics of binding. These scoring functions perform well when tested on similar targets to those in the training set, but fail to generalise to dissimilar targets. To test what a machine learning-based scoring function has learnt, input attribution—a technique for learning which features are important to a model when making a prediction on a particular data point—can be applied. If a model successfully learns something beyond dataset biases, attribution should give insight into the important binding interactions that are taking place. We built a machine learning-based scoring function that aimed to avoid the influence of bias via thorough train and test dataset filtering, and show that it achieves comparable performance on the CASF-2016 benchmark to other leading methods. We then use the CASF-2016 test set to perform attribution, and find that the bonds identified as important by PointVS, unlike those extracted from other scoring functions, have a high correlation with those found by a distance-based interaction profiler. We then show that attribution can be used to extract important binding pharmacophores from a given protein target when supplied with a number of bound structures. We use this information to perform fragment elaboration, and see improvements in docking scores compared to using structural information from a traditional, data-based approach. This not only provides definitive proof that the scoring function has learnt to identify some important binding interactions, but also constitutes the first deep learning-based method for extracting structural information from a target for molecule design.

Drug Discovery is a lengthy and costly process and has faced a period of declining productivity within the last two decades resulting in increasing importance of integrative data-driven approaches. In this paper, data mining and integration is leveraged to inspect target innovation trends in drug discovery. The study highlights protein families and classes that have received more attention and those that have just emerged in the scientific literature, thus highlighting novel opportunities for drug intervention. In order to delineate the evolution of target-driven research interest from a biological perspective, trends in biological process annotations from Gene Ontology and disease annotations from DisGeNET are captured. The analysis reveals an increasing interest in targets related to immune system processes, and a recurrent trend for targets involved in circulatory system processes. At the level of diseases, targets associated with cancer-related pathologies, intellectual disability, and schizophrenia are increasingly investigated in recent years. The methodology enables researchers to capture trends in research attention in target space at an early stage during the drug discovery process. Workflows, scripts, and data used in this study are publicly available from https://github.com/BZdrazil/Moving_Targets. An interactive web application allows the customized exploration of target, biological process, and disease trends (available at https://rguha.shinyapps.io/MovingTargets/).

The underlying structure of natural language is hierarchical; words combine into phrases, which in turn form clauses. An awareness of this hierarchical structure can aid machine learning models in performing many linguistic tasks. However, most such models just process text sequentially and there is no bias towards learning hierarchical structure encoded into their architecture. In this paper, we extend the recent transformer model (Vaswani et al., 2017) by enabling it to learn hierarchical representations. To achieve this, we adapt the ordering mechanism introduced in Shen et al., 2018, to the self-attention module of the transformer architecture. We train our new model on language modelling and then apply it to the task of unsupervised parsing. We achieve reasonable results on the freely available subset of the WSJ10 dataset with an F1-score of about 50%.

Protein pocket matching, or binding site comparison, is of importance in drug discovery. Identification of similar binding pockets can help guide efforts for hit-finding, understanding polypharmacology, and characterization of protein function. The design of pocket matching methods has traditionally involved much intuition and has employed a broad variety of algorithms and representations of the input protein structures. We regard the high heterogeneity of past work and the recent availability of large-scale benchmarks as an indicator that a data-driven approach may provide a new perspective. We propose DeeplyTough, a convolutional neural network that encodes a three-dimensional representation of protein pockets into descriptor vectors that may be compared efficiently in an alignment-free manner by computing pairwise Euclidean distances. The network is trained with supervision (i) to provide similar pockets with similar descriptors, (ii) to separate the descriptors of dissimilar pockets by a minimum margin, and (iii) to achieve robustness to nuisance variations. We evaluate our method using three large-scale benchmark datasets, on which it demonstrates excellent performance for held-out data coming from the training distribution and competitive performance when the trained network is required to generalize to datasets constructed independently. DeeplyTough is available at https://github.com/BenevolentAI/DeeplyTough.

Abstract Background: Nonalcoholic fatty liver disease (NAFLD) is a highly prevalent yet underdiagnosed and under-discussed disease. Given that NAFLD has not been explored sufficiently compared with other diseases, opportunities abound for scientists to discover new biomarkers (such as laboratory observations, current comorbidities, and behavioral descriptors) that can be linked to the development of conditions and complications that may develop at a later stage of the patient’s life.Methods: We analyzed IBM Explorys, a repository that contains electronic medical records (EMRs) of more than 60 million individuals. We used a classification algorithm that members of our group have previously validated to identify patients at a high probability for NAFLD. The algorithm identified more than 80,000 patients with a high probability for NAFLD who had at least 5 years of follow-up. We followed an unbiased approach for prediction modeling and applied standard statistical methods (such as logistic regression and bootstrapping) as well as Clinical Classifications Software (CCS) definitions to identify associations between a variety of covariates and disease outcomes.Results: Our methodology identified several thousand strongly statistically significant associations between covariates and outcomes in NAFLD. Most of the associations are known, but others may be new and require further investigation in subsequent studies.Conclusions: A discovery mechanism composed of standard statistical methods and applied on a large collection of EMRs confirmed known associations and identified potentially new associations that can act as biomarkers that might merit further research.

Extracting biomedical relations from large corpora of scientific documents is a challenging natural language processing task. Existing approaches usually focus on identifying a relation either in a single sentence (mention-level) or across an entire corpus (pair-level). In both cases, recent methods have achieved strong results by learning a point estimate to represent the relation; this is then used as the input to a relation classifier. However, the relation expressed in text between a pair of biomedical entities is often more complex than can be captured by a point estimate. To address this issue, we propose a latent variable model with an arbitrarily flexible distribution to represent the relation between an entity pair. Additionally, our model provides a unified architecture for both mention-level and pair-level relation extraction. We demonstrate that our model achieves results competitive with strong baselines for both tasks while having fewer parameters and being significantly faster to train. We make our code publicly available.