Your question is an interesting and common one, although most people ask about VC-fundable ideas vs boostrapped ones. We might do a project soon exploring opportunities in this area, but my feeling now is that there is a lot of interest in using ML / informatics to find new targets, find new ways to drug validated targets, and speed up lead optimization. the biggest bottleneck is creating scalable pipelines for getting high quality, annotated data. a good way to create these targeted pipelines is by starting with a targeted therapeutic / biological hypothesis and building the informatics platform from there
It is true that most startups are in biopharma / drug development, and that field has performed incredibly well the last five years. In the last 18 months or so, "bio-IT" has emerged as an area of significant interest. This is a pretty amorphous term (one VC described it as "a biotech startup whose CEO wears a hoodie, not a suit") that ranges in meaning from turning drug development into an engineering discipline [1] to using AI to identify novel targets or drug candidates (see recursion pharma as a good example of this type of company), to mining genomic and transcriptomic data in search of novel drug targets
a few things that might help frame the opportunity:
an MIT-affiliated venture creation firm released a good paper last year on the applications of AI to drug dev [2]
there was a post here a few days back on using AI to predict protein folding. tools like that would be quite useful
tools to predict tox / PK/PD of compounds is an area of interest
platforms to analyze various -omics data to identify disease associated pathways and targets are interesting
"virtual screening" tools have been an early area of research, either based on biochemistry / biophysics and structural data, or AI / ML approaches
however the holy grail is using informatics to find new validated targets. the failure of big late stage studies due to targets that worked in mice but not in humans is by far the biggest driver of the cost of drug r&d
my favorite example of using bioinformatics to find validated targets is the Regeneron Genetics Center. regeneron is a "big biotech" company with one of the best reputations for early stage r&d in the field. their approach is "lower tech" but quite clever and effective
RGC started with a clever therapeutic hypothesis that 1) gets the most "juice" out of each piece of data and 2) enables the targeted optimization of the data collection and analytics platform. the thesis is basically: drugs can really only target one molecule at a time, so you want to find single genetic mutations that are correlated with very dramatic clinical phenotypes (rather than a "signature" of a bunch of small effect sizes); rare phenotypes can inform generalizable clinical strategies (if one rare mutation results in, say, super low LDL in a tiny subset of patients, then you can probably target that mutation to reduce LDL in any patient); you need scale to find these rare and actionable mutations (so focus on lots of cheap WES vs higher resolution but more expensive WGS)
they dont use advanced AI or ML, but get cheap, high quality data through partnerships with health systems, have a high throughput, highly automated genomics core, and an informatics infrastructure that makes it easy to do the kinds of queries they need. many firms start with data and then search for therapeutic applications, but this has proved inefficient and ineffective, and RGC's approach stands out as very clever and scalable.
[1] https://a16z.com/2017/12/14/second-bio-fund/
[2] https://static1.squarespace.com/static/57b630ae9de4bb6c5c0b6...