Abstract
Ultrafast Shape Recognition (USR), along with its derivatives, are Ligand-Based Virtual Screening (LBVS) methods that condense 3-dimensional information about molecular shape, as well as other properties, into a small set of numeric descriptors. These can be used to efficiently compute a measure of similarity between pairs of molecules using a simple inverse Manhattan Distance metric. In this study we explore the use of suitable Machine Learning techniques that can be trained using USR descriptors, so as to improve the similarity detection of potential new leads. We use molecules from the Directory for Useful Decoys-Enhanced to construct machine learning models based on three different algorithms: Gaussian Mixture Models (GMMs), Isolation Forests and Artificial Neural Networks (ANNs). We train models based on full molecule conformer models, as well as the Lowest Energy Conformations (LECs) only. We also investigate the performance of our models when trained on smaller datasets so as to model virtual screening scenarios when only a small number of actives are known a priori. Our results indicate significant performance gains over a state of the art USR-derived method, ElectroShape 5D, with GMMs obtaining a mean performance up to 430% better than that of ElectroShape 5D in terms of Enrichment Factor with a maximum improvement of up to 940%. Additionally, we demonstrate that our models are capable of maintaining their performance, in terms of enrichment factor, within 10% of the mean as the size of the training dataset is successively reduced. Furthermore, we also demonstrate that running times for retrospective screening using the machine learning models we selected are faster than standard USR, on average by a factor of 10, including the time required for training. Our results show that machine learning techniques can significantly improve the virtual screening performance and efficiency of the USR family of methods.
Highlights
The discovery and development of a new drug is a timeconsuming process that can take 14 years to complete successfully, incurring a cost of about 2.5 billion US dollars (DiMasi et al, 2016)
To the best of our knowledge, this research project constitutes the first study to explore the viability of several machine learning algorithms in their application to Ligand-Based Virtual Screening (LBVS) using Ultrafast Shape Recognition (USR) and USRlike descriptors
We have demonstrated the utility of applying machine learning methods to the LBVS scenario when using USR-like descriptors, managing to obtain significant performance improvements over both the USR and the ElectroShape 5D algorithms using the Gaussian Mixture Model (GMM), Isolation Forest and Artificial Neural Network (ANN) algorithms
Summary
The discovery and development of a new drug is a timeconsuming process that can take 14 years to complete successfully, incurring a cost of about 2.5 billion US dollars (DiMasi et al, 2016). Virtual Screening (VS) is a search approach that leverages electronic databases of chemical compounds and modern computing resources to streamline this process. The aim of this process is to computationally pre-screen molecules to find those that are most likely to exhibit affinity for binding to a given target protein. In this way, laboratory time and resources associated with High Throughput Screening (HTS) can be drastically reduced by preferentially testing only the compounds that are more likely to become successful leads (Leach and Gillet, 2007). The top molecules from this list, i.e. the most similar to the bioactive one, are moved forward for physical testing
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
