Artificial intelligence-driven approach for patient-focused drug development.

Patients' increasing digital participation provides an opportunity to pursue patient-centric research and drug development by understanding their needs. Social media has proven to be one of the most useful data sources when it comes to understanding a company's potential audience to drive more targeted impact. Navigating through an ocean of information is a tedious task where techniques such as artificial intelligence and text analytics have proven effective in identifying relevant posts for healthcare business questions. Here, we present an enterprise-ready, scalable solution demonstrating the feasibility and utility of social media-based patient experience data for use in research and development through capturing and assessing patient experiences and expectations on disease, treatment options, and unmet needs while creating a playbook for roll-out to other indications and therapeutic areas.

Weakly Supervised Learning for Categorization of Medical Inquiries for Customer Service Effectiveness.

With the growing unstructured data in healthcare and pharmaceutical, there has been a drastic adoption of natural language processing for generating actionable insights from text data sources. One of the key areas of our exploration is the Medical Information function within our organization. We receive a significant amount of medical information inquires in the form of unstructured text. An enterprise-level solution must deal with medical information interactions via multiple communication channels which are always nuanced with a variety of keywords and emotions that are unique to the pharmaceutical industry. There is a strong need for an effective solution to leverage the contextual knowledge of the medical information business along with digital tenants of natural language processing (NLP) and machine learning to build an automated and scalable process that generates real-time insights on conversation categories. The traditional supervised learning methods rely on a huge set of manually labeled training data and this dataset is difficult to attain due to high labeling costs. Thus, the solution is incomplete without its ability to self-learn and improve. This necessitates techniques to automatically build relevant training data using a weakly supervised approach from textual inquiries across consumers, healthcare professionals, sales, and service providers. The solution has two fundamental layers of NLP and machine learning. The first layer leverages heuristics and knowledgebase to identify the potential categories and build an annotated training data. The second layer, based on machine learning and deep learning, utilizes the training data generated using the heuristic approach for identifying categories and sub-categories associated with verbatim. Here, we present a novel approach harnessing the power of weakly supervised learning combined with multi-class classification for improved categorization of medical information inquiries.

Identification of pharmacodynamic biomarker hypotheses through literature analysis with IBM Watson.

BACKGROUND: Pharmacodynamic biomarkers are becoming increasingly valuable for assessing drug activity and target modulation in clinical trials. However, identifying quality biomarkers is challenging due to the increasing volume and heterogeneity of relevant data describing the biological networks that underlie disease mechanisms. A biological pathway network typically includes entities (e.g. genes, proteins and chemicals/drugs) as well as the relationships between these and is typically curated or mined from structured databases and textual co-occurrence data. We propose a hybrid Natural Language Processing and directed relationships-based network analysis approach using IBM Watson for Drug Discovery to rank all human genes and identify potential candidate biomarkers, requiring only an initial determination of a specific target-disease relationship. METHODS: Through natural language processing of scientific literature, Watson for Drug Discovery creates a network of semantic relationships between biological concepts such as genes, drugs, and diseases. Using Bruton's tyrosine kinase as a case study, Watson for Drug Discovery's automatically extracted relationship network was compared with a prominent manually curated physical interaction network. Additionally, potential biomarkers for Bruton's tyrosine kinase inhibition were predicted using a matrix factorization approach and subsequently compared with expert-generated biomarkers. RESULTS: Watson's natural language processing generated a relationship network matching 55 (86%) genes upstream of BTK and 98 (95%) genes downstream of Bruton's tyrosine kinase in a prominent manually curated physical interaction network. Matrix factorization analysis predicted 11 of 13 genes identified by Merck subject matter experts in the top 20% of Watson for Drug Discovery's 13,595 ranked genes, with 7 in the top 5%. CONCLUSION: Taken together, these results suggest that Watson for Drug Discovery's automatic relationship network identifies the majority of upstream and downstream genes in biological pathway networks and can be used to help with the identification and prioritization of pharmacodynamic biomarker evaluation, accelerating the early phases of disease hypothesis generation.

Automatic detection of adverse events to predict drug label changes using text and data mining techniques.

PURPOSE: The aim of this study was to assess the impact of automatically detected adverse event signals from text and open-source data on the prediction of drug label changes. METHODS: Open-source adverse effect data were collected from FAERS, Yellow Cards and SIDER databases. A shallow linguistic relation extraction system (JSRE) was applied for extraction of adverse effects from MEDLINE case reports. Statistical approach was applied on the extracted datasets for signal detection and subsequent prediction of label changes issued for 29 drugs by the UK Regulatory Authority in 2009. RESULTS: 76% of drug label changes were automatically predicted. Out of these, 6% of drug label changes were detected only by text mining. JSRE enabled precise identification of four adverse drug events from MEDLINE that were undetectable otherwise. CONCLUSIONS: Changes in drug labels can be predicted automatically using data and text mining techniques. Text mining technology is mature and well-placed to support the pharmacovigilance tasks.

'HypothesisFinder:' a strategy for the detection of speculative statements in scientific text.

Speculative statements communicating experimental findings are frequently found in scientific articles, and their purpose is to provide an impetus for further investigations into the given topic. Automated recognition of speculative statements in scientific text has gained interest in recent years as systematic analysis of such statements could transform speculative thoughts into testable hypotheses. We describe here a pattern matching approach for the detection of speculative statements in scientific text that uses a dictionary of speculative patterns to classify sentences as hypothetical. To demonstrate the practical utility of our approach, we applied it to the domain of Alzheimer's disease and showed that our automated approach captures a wide spectrum of scientific speculations on Alzheimer's disease. Subsequent exploration of derived hypothetical knowledge leads to generation of a coherent overview on emerging knowledge niches, and can thus provide added value to ongoing research activities.

Extraction of potential adverse drug events from medical case reports.

: The sheer amount of information about potential adverse drug events published in medical case reports pose major challenges for drug safety experts to perform timely monitoring. Efficient strategies for identification and extraction of information about potential adverse drug events from free-text resources are needed to support pharmacovigilance research and pharmaceutical decision making. Therefore, this work focusses on the adaptation of a machine learning-based system for the identification and extraction of potential adverse drug event relations from MEDLINE case reports. It relies on a high quality corpus that was manually annotated using an ontology-driven methodology. Qualitative evaluation of the system showed robust results. An experiment with large scale relation extraction from MEDLINE delivered under-identified potential adverse drug events not reported in drug monographs. Overall, this approach provides a scalable auto-assistance platform for drug safety professionals to automatically collect potential adverse drug events communicated as free-text data.

Development of a benchmark corpus to support the automatic extraction of drug-related adverse effects from medical case reports.

A significant amount of information about drug-related safety issues such as adverse effects are published in medical case reports that can only be explored by human readers due to their unstructured nature. The work presented here aims at generating a systematically annotated corpus that can support the development and validation of methods for the automatic extraction of drug-related adverse effects from medical case reports. The documents are systematically double annotated in various rounds to ensure consistent annotations. The annotated documents are finally harmonized to generate representative consensus annotations. In order to demonstrate an example use case scenario, the corpus was employed to train and validate models for the classification of informative against the non-informative sentences. A Maximum Entropy classifier trained with simple features and evaluated by 10-fold cross-validation resulted in the F1 score of 0.70 indicating a potential useful application of the corpus.

Concept-based semi-automatic classification of drugs.

The anatomical therapeutic chemical (ATC) classification system maintained by the World Health Organization provides a global standard for the classification of medical substances and serves as a source for drug repurposing research. Nevertheless, it lacks several drugs that are major players in the global drug market. In order to establish classifications for yet unclassified drugs, this paper presents a newly developed approach based on a combination of information extraction (IE) and machine learning (ML) techniques. Most of the information about drugs is published in the scientific articles. Therefore, an IE-based framework is employed to extract terms from free text that express drug's chemical, pharmacological, therapeutic, and systemic effects. The extracted terms are used as features within a ML framework to predict putative ATC class labels for unclassified drugs. The system was tested on a portion of ATC containing drugs with an indication on the cardiovascular system. The class prediction turned out to be successful with the best predictive accuracy of 89.47% validated by a 100-fold bootstrapping of the training set and an accuracy of 77.12% on an independent test set. The presented concept-based classification system outperformed state-of-the-art classification methods based on chemical structure properties.