My overarching goals are to expand medical reasoning beyond the limits of human faculties and to use unstructured data from the Internet to identify spatiotemporal patterns of substance use, disease exposure, and disease manifestation. To this end, I use natural language processing, machine learning, and artificial intelligence to create computational versions of medical and scientific knowledge to allow computers to reason about medicine as experienced clinicians might.
Current questions of interest include: (i) What are the spatiotemporal patterns of use of novel psychoactive substances?, (ii) Can we infer nover signal transduction pathways or even therapeutics from online discussions about novel psychoactive substance use?, and (iii) What representations of knowledge to we need to do (i) and (ii) at scale?
Novel psychoactive substances include synthetic opioids such as fentanyl or carfentanil, substituted phenethylamines, such as ‘‘bath salts’’, and synthetic cannabinoid receptor agonists (‘‘synthetic marijuana’’). The rate of emergence of these substances has outstripped the ability of traditional means of public health surveillance to identify the substances and presentations or toxicities of each substance. The central finding of my work is that patterns of use of novel psychoactive substances can be estimated from social media (YouTube, Twitter, online discussion fora. I demonstrated that this information was corroborated by accepted sources of information on substance use (National Survey on Drug Usage & Health) for opioids and dextromethorphan (an ingredient in cough syrup that in large amounts has euphoric and dissociative effects). These studies opened the door to using social media to discover new methods of abuse (e.g. abusing Immodium (loperamide) to prevent opioid withdrawal) and better understand the structure-function relationship in derivatives of MDMA (Ecstasy) using user reports submitted online.
Social Media can also provide a starting point for identifying compounds that lead to new therapeutics and help us better understand how the brain functions.
The volume of knowledge about drugs of abuse, including novel psychoactive substances, outstrips human processing capability. The acumen of experience clinicians, nevertheless, identifies patterns and infers knowledge from data in ways that current approaches to artificial intelligence cannot replicate. A digital representation of medical decision-making could allow computers to reason at the scale of Big Data with the acumen of an experienced clinician. The central finding of my work in this area is that ontologies can create a digital representation of medical reasoning about opioid use disorders.