Mining pharmacogenomic information from drug labeling using FDALabel database for advancing precision medicine

It is known that drug response can have significant interpatient variability. The relationship between drug response and genetic makeup of an individual/population, studied in the field of pharmacogenomics, is rapidly accelerating advancements towards precision medicine. The US FDA has included pharmacogenomic information in the labeling of approved drug products to improve drug safety and efficacy. However, this resource has not been fully utilized by the research community and its applications in precision medicine have been limited. In this research, we present an overview of the large amount of pharmacogenomic information contained in drug labeling using FDALabel database and propose a novel classification scheme that enhances the understanding and utility of this information. FDALabel is a web-based application that allows users to perform customizable searches of about 95,000 labeling documents that include human prescription and over-the-counter (OTC) drugs. Using a set of 62 biomarkers obtained from a public FDA resource website, we queried FDALabel database and identified 225 drugs with biomarker information in their labeling and 289 drug-biomarker pairs. We then classified these drug-biomarker pairs into 4 categories of ADRs (predicting patients at risk for adverse drug reactions), Dose-related (guiding dosage adjustments for optimal drug effect), Indication (predicting responders by biomarkers), and Informative (descriptive information only). An analysis of the relationship between drugs and biomarkers revealed that the most frequently observed biomarkers in drug labeling are CYP2D6, G6PD, and CYP2C19 and the most frequently observed therapeutic areas are oncology, psychiatry, and infectious diseases. The drug-biomarker pair distribution across the 4-way classification scheme revealed that 62% of oncology drugs require proper patient selection before treatment and 75% of psychiatry drugs require dosage adjustment for patients of different genotypes. In summary, the classification scheme proposed here is found to be effective and aligns well with regulatory science applications in monitoring drug safety and efficacy as well as the goal of precision medicine in selecting the right drug at the right dose for the right patient.

ToxML data redaction tool for chemical/ toxicity data sharing with external collaborators

The Office of Food Additive Safety (OFAS) in CFSAN periodically shares non-confidential data with external collaborators from industry, international consortiums, and government organizations to facilitate the understanding of toxicity of chemicals and to support the development of models for predicting chemical toxicity. In order to preserve the confidentiality of data, it is essential that the data be scrutinized for any Confidential Commercial Information (CCI) such as trade names, before sharing it with collaborators. Current methods of data sharing involve manual curation of data in an Excel sheet line by line and require the collaborators to parse these data into their preferred format, thus introducing the possibility of parsing errors. We have developed a data redaction tool that automates the process of data screening and allows sharing of redacted data in xml format. This data redaction tool is designed to compile information from ToxML files, an open data exchange standard for sharing toxicological and related data. The tool takes multiple ToxML files as input, checks for the presence of trade names and replaces them with a generic phrase, redacts fields containing confidential information that are marked for redaction, and generates a single compiled ToxML file as output. The tool also saves a csv file as output with records of all fields where trade names were found and replaced. This preserves transparency of the tool’s automated process, thereby ensuring proper management of CCI.

Evaluating performance of chemical fingerprinting methods and machine learning algorithms for in silico prediction of Ames mutagenicity

The Office of Food Additive Safety (OFAS) at U.S. FDA’s Center for Food Safety and Applied Nutrition is responsible for ensuring the safety of all food additives used in the United States. Current research efforts at OFAS focus on building in-house mutagenicity and carcinogenicity predictive models with high prediction accuracy for food related chemicals. In this research, we present an evaluation of different chemical fingerprinting methods and machine learning algorithms available in the public domain and compare their performance for in silico prediction of Ames mutagenicity. We evaluated six fingerprinting methods; MACCS keys, RDKit, Circular, ToxPrint, PubChem, and Atom pairs, and six machine learning algorithms; k-Nearest Neighbors, Decision Trees, Random Forest, Artificial Neural Networks, Support Vector Machines, and Naïve Bayes. QSAR models were developed using all combinations of fingerprints and machine learning algorithms and performance metrics were calculated using the Hansen benchmark dataset. Combinations of knowledge-enriched fingerprints and deep learning algorithms were found to give the best performing models for Ames mutagenicity. Some of these models were then evaluated against empirical data on food related chemicals in the OFAS food additive knowledgebase called CERES. These models were found to give good overall predictive performance and high accuracy in predicting the non-mutagenic compounds. More research is needed to improve the prediction of mutagenic compounds in the food related chemical space.

Structure-based in silico modeling of chemically-induced toxicity

Evaluating the potential toxicity of chemical compounds is an important step in the development of most new products, ranging from drugs and food additives to packaging materials and cosmetics. Data on the toxicity of these compounds are often not available because a thorough experimental study of every candidate compound can be too expensive and time consuming to be feasible. It is thus necessary to develop computational models that can help to screen candidate compounds by predicting their toxicity using chemical structural information. These models are developed with the help of chemical informatics methods that allow structural information to be captured using chemical descriptors and used for modeling complex endpoints such as toxicity using machine learning techniques. In this research, we propose to generate novel chemical descriptors and develop a new machine learning technique for modeling. The novel descriptors are generated by fragmenting chemical structures into linear paths of connected atoms and annotating these with atom-based features such as atom identity and partial charge. These features provide flexibility in exploring different degrees of structural details from the same chemical fragment. These annotated chemical fragments are then used for modeling toxicity using a statistical method called as Markov chain analysis. The performance of this method is evaluated using previous experimental toxicity data on skin sensitization and Ames mutagenicity. The preliminary results show that the new method is able to capture more meaningful descriptors and yield higher prediction accuracies as compared to other methods. We plan to refine this method further by including additional annotation features, developing more complex Markov models, and exploring effects of co-occurring fragments. The goal of this refinement is to make the method more robust and increase its predictive performance based on validation against known datasets. This research will help to reduce the time and money required to bring new products to market by assisting researchers in the lab to prioritize experimental tests. This research may also lead to the discovery of new structural alerts related to chemical toxicity, thereby increasing our understanding of the mechanisms underlying these complex phenomena.

Check this out!