Influence Relevance Voting: An Accurate And Interpretable Virtual High Throughput Screening Method

Abstract

Given activity training data from Hight-Throughput Screening (HTS) experiments, virtual HighThroughput Screening (vHTS) methods aim to predict in silico the activity of untested chemicals. We present a novel method, the Influence Relevance Voter (IRV), specifically tailored for the vHTS task. The IRV is a low-parameter neural network which refines a k-nearest neighbor classifier by non-linearly combining the influences of a chemical's neighbors in the training set. Influences are decomposed, also non-linearly, into a relevance component and a vote component.The IRV is benchmarked using the data and rules of two large, open, competitions, and its performance compared to the performance of other participating methods, as well as of an in-house Support Vector Machine (SVM) method. On these benchmark datasets, IRV achieves state-of-theart results, comparable to the SVM in one case, and significantly better than the SVM in the other, retrieving three times as many actives in the top 1% of its prediction-sorted list.The IRV presents several other important advantages over SVMs and other methods: (1) the output predictions have a probabilistic semantic; (2) the underlying inferences are interpretable; (3) the training time is very short, on the order of minutes even for very large data sets; (4) the risk of overfitting is minimal, due to the small number of free parameters; and (5) additional information can easily be incorporated into the IRV architecture. Combined with its performance, these qualities make the IRV particularly well suited for vHTS.Virtual High-Throughput Screening (vHTS) is the cost-effective, in silico complement of experimental HTS. A vHTS algorithm uses data from HTS experiments to predict the activity of new sets of compounds in silico. Although vHTS is sometimes cast as a classification task, it is more appropriately described as a ranking task, where the goal is to rank additional compounds, such that active compounds are close to the the top of the prediction-sorted list as possible. The experiments required to verify a hit are expensive, so it is critical that true actives be recognized as early as possible. Accurately ordering actives by their degree of activity, however, is not critical. The vHTS task, therefore, differs from the `ranking' task of the machine learning literature, in that the goal is not to precisely order the chemicals in relation to each other, but rather to globally rank as many actives as possible above the bulk of the inactives. Furthermore, proper vHTS training data for the ranking task, is often unavailable. E-mail: sswamida@ics.uci NIH-PA Author ManuscriptNIH-PA Author Manuscript NIH-PA Author ManuscriptAn important algorithm proposed for vHTS is the k-Nearest Neighbor (kNN) classifier, a nonparametric method which has been shown effective in a number of other problems. 1,2 In the kNN approach, each new data point is classified by integrating information from its neighborhood in the training set in a very simple way. Specifically, a new data point is assigne...

Molecular graph convolutions: moving beyond fingerprints

“…We also trained graph convolution models on some additional datasets in order to compare to the “neural fingerprints” (NFP) of Duvenaud et al [9] and the influence relevance voter (IRV) method of Swamidass et al [36] (see “Comparisons to other methods” section). Table 5 compares graph convolution models to published results on these datasets under similar cross-validation conditions.…”

“…In this work, we emphasize their application to small molecules—undirected graphs of atoms connected by bonds—for virtual screening. Starting from simple descriptions of atoms, bonds between atoms, and pairwise relationships in a molecular graph, we have demonstrated performance that is comparable to state of the art multitask neural networks trained on traditional molecular fingerprint representations, as well as alternative methods including “neural fingerprints” [9] and influence relevance voter [36]. …”

Machine learning and similarity-based virtual screening techniques

“…The approach assesses the influence of structural neighbors of a compound on its classification as active or inactive. In initial benchmark investigations, IRV compound recall was at least comparable to SVMs [21]. Different from many other machine-learning approaches, IRV models are chemically interpretable.…”

“…As a general caveat, it is often difficult to understand whether complex LBVS protocols are truly required to identify active compounds or which of the components might have contributed most. Furthermore, as an exemplary new methodology, the influence relevance voter (IRV) has been introduced that combines a low-level neural network with a k-nearest neighbor classifier [21]. The approach assesses the influence of structural neighbors of a compound on its classification as active or inactive.…”

A CROC stronger than ROC: measuring, visualizing and optimizing early retrieval

“…For the SVM algorithm, it is well known that the Tanimoto similarity satisfies Mercer's condition (Swamidass et al, 2005a), and therefore it defines a kernel that can be used in an SVM for classification (see Azencott et al, 2007;Mahé et al, 2006;Swamidass et al, 2005b, for other related kernels). Finally, we use the IRV algorithm (Swamidass et al, 2009), which can be viewed as a refinement of kNN. At a high-level, the IRV is defined by a preprocessing step, during which all the neighbors-as defined by the Tanimoto similarity metric-of a test molecule are identified, and a processing step during which information from each neighbor is fed into a neural network to produce a prediction.…”