PrivacyNet: Semi-Adversarial Networks for Multi-attribute Face Privacy.

Recent research has established the possibility of deducing soft-biometric attributes such as age, gender and race from an individual's face image with high accuracy. However, this raises privacy concerns, especially when face images collected for biometric recognition purposes are used for attribute analysis without the person's consent. To address this problem, we develop a technique for imparting soft biometric privacy to face images via an image perturbation methodology. The image perturbation is undertaken using a GAN-based Semi-Adversarial Network (SAN) - referred to as PrivacyNet - that modifies an input face image such that it can be used by a face matcher for matching purposes but cannot be reliably used by an attribute classifier. Further, PrivacyNet allows a person to choose specific attributes that have to be obfuscated in the input face images (e.g., age and race), while allowing for other types of attributes to be extracted (e.g., gender). Extensive experiments using multiple face matchers, multiple age/gender/race classifiers, and multiple face datasets demonstrate the generalizability of the proposed multi-attribute privacy enhancing method across multiple face and attribute classifiers.

Machine learning and AI-based approaches for bioactive ligand discovery and GPCR-ligand recognition.

In the last decade, machine learning and artificial intelligence applications have received a significant boost in performance and attention in both academic research and industry. The success behind most of the recent state-of-the-art methods can be attributed to the latest developments in deep learning. When applied to various scientific domains that are concerned with the processing of non-tabular data, for example, image or text, deep learning has been shown to outperform not only conventional machine learning but also highly specialized tools developed by domain experts. This review aims to summarize AI-based research for GPCR bioactive ligand discovery with a particular focus on the most recent achievements and research trends. To make this article accessible to a broad audience of computational scientists, we provide instructive explanations of the underlying methodology, including overviews of the most commonly used deep learning architectures and feature representations of molecular data. We highlight the latest AI-based research that has led to the successful discovery of GPCR bioactive ligands. However, an equal focus of this review is on the discussion of machine learning-based technology that has been applied to ligand discovery in general and has the potential to pave the way for successful GPCR bioactive ligand discovery in the future. This review concludes with a brief outlook highlighting the recent research trends in deep learning, such as active learning and semi-supervised learning, which have great potential for advancing bioactive ligand discovery.

Machine Learning to Identify Flexibility Signatures of Class A GPCR Inhibition.

We show that machine learning can pinpoint features distinguishing inactive from active states in proteins, in particular identifying key ligand binding site flexibility transitions in GPCRs that are triggered by biologically active ligands. Our analysis was performed on the helical segments and loops in 18 inactive and 9 active class A G protein-coupled receptors (GPCRs). These three-dimensional (3D) structures were determined in complex with ligands. However, considering the flexible versus rigid state identified by graph-theoretic ProFlex rigidity analysis for each helix and loop segment with the ligand removed, followed by feature selection and k-nearest neighbor classification, was sufficient to identify four segments surrounding the ligand binding site whose flexibility/rigidity accurately predicts whether a GPCR is in an active or inactive state. GPCRs bound to inhibitors were similar in their pattern of flexible versus rigid regions, whereas agonist-bound GPCRs were more flexible and diverse. This new ligand-proximal flexibility signature of GPCR activity was identified without knowledge of the ligand binding mode or previously defined switch regions, while being adjacent to the known transmission switch. Following this proof of concept, the ProFlex flexibility analysis coupled with pattern recognition and activity classification may be useful for predicting whether newly designed ligands behave as activators or inhibitors in protein families in general, based on the pattern of flexibility they induce in the protein.

Automated discovery of GPCR bioactive ligands.

While G-protein-coupled receptors (GPCRs) constitute the largest class of membrane proteins, structures and endogenous ligands of a large portion of GPCRs remain unknown. Because of the involvement of GPCRs in various signaling pathways and physiological roles, the identification of endogenous ligands as well as designing novel drugs is of high interest to the research and medical communities. Along with highlighting the recent advances in structure-based ligand discovery, including docking and molecular dynamics, this article focuses on the latest advances for automating the discovery of bioactive ligands using machine learning. Machine learning is centered around the development and applications of algorithms that can learn from data automatically. Such an approach offers immense opportunities for bioactivity prediction as well as quantitative structure-activity relationship studies. This review describes the most recent and successful applications of machine learning for bioactive ligand discovery, concluding with an outlook on deep learning methods that are capable of automatically extracting salient information from structural data as a promising future direction for rapid and efficient bioactive ligand discovery.

Automated Inference of Chemical Discriminants of Biological Activity.

Ligand-based virtual screening has become a standard technique for the efficient discovery of bioactive small molecules. Following assays to determine the activity of compounds selected by virtual screening, or other approaches in which dozens to thousands of molecules have been tested, machine learning techniques make it straightforward to discover the patterns of chemical groups that correlate with the desired biological activity. Defining the chemical features that generate activity can be used to guide the selection of molecules for subsequent rounds of screening and assaying, as well as help design new, more active molecules for organic synthesis.The quantitative structure-activity relationship machine learning protocols we describe here, using decision trees, random forests, and sequential feature selection, take as input the chemical structure of a single, known active small molecule (e.g., an inhibitor, agonist, or substrate) for comparison with the structure of each tested molecule. Knowledge of the atomic structure of the protein target and its interactions with the active compound are not required. These protocols can be modified and applied to any data set that consists of a series of measured structural, chemical, or other features for each tested molecule, along with the experimentally measured value of the response variable you would like to predict or optimize for your project, for instance, inhibitory activity in a biological assay or ΔGbinding. To illustrate the use of different machine learning algorithms, we step through the analysis of a dataset of inhibitor candidates from virtual screening that were tested recently for their ability to inhibit GPCR-mediated signaling in a vertebrate.

Protein-ligand interfaces are polarized: discovery of a strong trend for intermolecular hydrogen bonds to favor donors on the protein side with implications for predicting and designing ligand complexes.

Understanding how proteins encode ligand specificity is fascinating and similar in importance to deciphering the genetic code. For protein-ligand recognition, the combination of an almost infinite variety of interfacial shapes and patterns of chemical groups makes the problem especially challenging. Here we analyze data across non-homologous proteins in complex with small biological ligands to address observations made in our inhibitor discovery projects: that proteins favor donating H-bonds to ligands and avoid using groups with both H-bond donor and acceptor capacity. The resulting clear and significant chemical group matching preferences elucidate the code for protein-native ligand binding, similar to the dominant patterns found in nucleic acid base-pairing. On average, 90% of the keto and carboxylate oxygens occurring in the biological ligands formed direct H-bonds to the protein. A two-fold preference was found for protein atoms to act as H-bond donors and ligand atoms to act as acceptors, and 76% of all intermolecular H-bonds involved an amine donor. Together, the tight chemical and geometric constraints associated with satisfying donor groups generate a hydrogen-bonding lock that can be matched only by ligands bearing the right acceptor-rich key. Measuring an index of H-bond preference based on the observed chemical trends proved sufficient to predict other protein-ligand complexes and can be used to guide molecular design. The resulting Hbind and Protein Recognition Index software packages are being made available for rigorously defining intermolecular H-bonds and measuring the extent to which H-bonding patterns in a given complex match the preference key.

Enabling the hypothesis-driven prioritization of ligand candidates in big databases: Screenlamp and its application to GPCR inhibitor discovery for invasive species control.

While the advantage of screening vast databases of molecules to cover greater molecular diversity is often mentioned, in reality, only a few studies have been published demonstrating inhibitor discovery by screening more than a million compounds for features that mimic a known three-dimensional (3D) ligand. Two factors contribute: the general difficulty of discovering potent inhibitors, and the lack of free, user-friendly software to incorporate project-specific knowledge and user hypotheses into 3D ligand-based screening. The Screenlamp modular toolkit presented here was developed with these needs in mind. We show Screenlamp's ability to screen more than 12 million commercially available molecules and identify potent in vivo inhibitors of a G protein-coupled bile acid receptor within the first year of a discovery project. This pheromone receptor governs sea lamprey reproductive behavior, and to our knowledge, this project is the first to establish the efficacy of computational screening in discovering lead compounds for aquatic invasive species control. Significant enhancement in activity came from selecting compounds based on one of the hypotheses: that matching two distal oxygen groups in the 3D structure of the pheromone is crucial for activity. Six of the 15 most active compounds met these criteria. A second hypothesis-that presence of an alkyl sulfate side chain results in high activity-identified another 6 compounds in the top 10, demonstrating the significant benefits of hypothesis-driven screening.

Detecting the native ligand orientation by interfacial rigidity: SiteInterlock.

Understanding the physical attributes of protein-ligand interfaces, the source of most biological activity, is a fundamental problem in biophysics. Knowing the characteristic features of interfaces also enables the design of molecules with potent and selective interactions. Prediction of native protein-ligand interactions has traditionally focused on the development of physics-based potential energy functions, empirical scoring functions that are fit to binding data, and knowledge-based potentials that assess the likelihood of pairwise interactions. Here we explore a new approach, testing the hypothesis that protein-ligand binding results in computationally detectable rigidification of the protein-ligand interface. Our SiteInterlock approach uses rigidity theory to efficiently measure the relative interfacial rigidity of a series of small-molecule ligand orientations and conformations for a number of protein complexes. In the majority of cases, SiteInterlock detects a near-native binding mode as being the most rigid, with particularly robust performance relative to other methods when the ligand-free conformation of the protein is provided. The interfacial rigidification of both the protein and ligand prove to be important characteristics of the native binding mode. This measure of rigidity is also sensitive to the spatial coupling of interactions and bond-rotational degrees of freedom in the interface. While the predictive performance of SiteInterlock is competitive with the best of the five other scoring functions tested, its measure of rigidity encompasses cooperative rather than just additive binding interactions, providing novel information for detecting native-like complexes. SiteInterlock shows special strength in enhancing the prediction of native complexes by ruling out inaccurate poses. Proteins 2016; 84:1888-1901. © 2016 Wiley Periodicals, Inc.