Webinar: Is Genomic Medicine Ready for A.I.?

Webinar Q & A

When will Mastermind have indexed a complete set (including supplemental material) of references so that other methods for manual literature searches will be obsolete?

We already have the most comprehensive corpus of this information available, and have rendered most external methods unnecessary with (in many circumstances) more than 10x the content of these previous gold-standard resources.

It is not possible to say that every conceivable reference will be included, but we are approaching the asymptote for full-text content repletion. We are quickly expanding our coverage of supplemental materials, although we only just began this month (September 2018).

We are able to directly prioritize the content that is of value to our users, and we welcome insight from them – either at the individual article level or otherwise with a focus on specific genes or diseases.

Is there a list of journals from which Mastermind indexes articles?

We are focused more on the content of references as opposed to the source. However, our current catalogue comprises more than 15k different titles.

If it is accessible to you as a clinician or researcher and it has genetic or genomic content, it is sought after by our content acquisition team for inclusion in the Mastermind index.

How does Mastermind adapt to deviations in variant nomenclature – for example, use of a unusual nomenclature or legacy nomenclature?

Mastermind recognizes patterns of variant mentions or deviations from accepted nomenclatures in order to ensure maximal sensitivity. This is necessary for non-systematic deviations or colloquial nomenclature, such as deltaF508 which is understood to refer to CFTR’s frequently cited p.F508del variation.

Much of the first two years of our operation had been focused on this very problem, so we have increasingly fewer and fewer examples to correct. Once we or one of our users identifies an issue, we are able to correct for this pattern in our code and re-index our corpus of references, and can do so in real-time.

There are also systematic changes such as positional shifts in transcripts that can occur over revisions to the reference genome, which we are actively working on solving with an automated approach to make it less prone to errors or oversight.

How does Mastermind resolve gene mentions that have overlap with other medical acronyms or otherwise English words – for example the MET gene?

There are a variety of techniques that we use to address these issues. We use a variety of techniques to determine the likelihood that a matched string is referring to the gene as opposed to an English word or medical acronym. As an example, various context cues including mentions of genetic and related keywords are used to quantitate the probability that the sentence containing the match is referring to a gene, which significantly improves the artifact profile in these rare circumstances where a gene symbol or synonym causes such false positives.

In the user interface, the Mastermind results are delivered in Sensitivity Mode to ensure first and foremost that no meaningful results are missed. Our current R+D efforts are focused on enhancing the specificity of these results. In our service work and APIs, we are able to stratify the results using these internal tools and techniques, with all final deliverables examined manually to ensure a high quality result.

How much do you think the need for evidence in clinical contexts will shape approaches to AI in the industry?

Genomenon feels that human assessment of AI-powered results will be required indefinitely in clinical settings, and therefore that putting an emphasis on the primacy of evidence and traceability of the results is critical to ensure successful application of AI for patient care in genomics. It is unclear how much this necessity will shape the AI activities of other groups, but it certainly is of utmost importance in our approaches.

Does your opinion of AI lean more toward AI-optimist Google CEO Sundar Pichai or AI-skeptic Stephen Hawking?

I’m not sure their opinions are necessarily opposed. You could say the same about the mechanisation that came with the Industrial Revolution. It’s been massively beneficial to the overall quality of life for billions of people over the past couple centuries, but now we’re realizing how some of our creations have adversely affected our planet, and seeking to correct course. The question really is, are we sufficiently aware of the potential consequences of AI to be able to make the right decisions? The tension between optimism and pessimism is a healthy and necessary process for any new revolution, especially as each successive technological revolution makes more sweeping and consequential changes to human societies (be they good or bad) than the last.