User interfaces for biology have drastically improved over the last 10 years.
Domain-specific tools like genome browsers, protein viewers, or phylogenetic explorers [1-3] almost all look and feel a lot better than they did in 2012.
The biggest exception here is UCSC Genome Browser, which has an old-school design and web technology stack. That said, it's steadily added features over the years, has substantially sleekened UX in its periphery, and remains widely used.
There are also bespoke visual design resources for biology applications that are good and getting better, like BioRender and PhyloPic [4-5]. There are multi-tiered packages like Dash Bio that wrap biology components together [6]. There's a Blender biology community, too!
I'm more interested in read speed than write speed. I have about 2 MB of data that I fetch, parse and transform into a nested object for easy look-up by various types of keys. It consists of 6 other objects, and I'd guess it's < 50 MB in total size.
In my brief experiment, it was 12% faster to read from the web Cache API [1], re-parse and re-transform that nested object than to read the fully transformed object using IndexedDB via idb-keyval [2]. That surprised me! I went on to learn that IndexedDB does a structured clone as part of such reads, which I suspect is the main cause of slowness in my use case.
Related commits to reproduce that finding are in [3], specifically [4].
WikiPathways supports advanced queries via their SPARQL API and UI. See [1] and [2]. I find WikiPathways nice because it lets logged-in users create and edit pathways, with a low barrier to entry.
I've been building a way to find related genes using biochemical pathways [3]. The source code linked there includes practical examples for fetching information on genes in those pathways, which you rightly note is needed for something compelling. That and other code there might help spark ideas for you on how to glue together various biochemistry and molecular biology APIs to achieve your vision.
I'm currently working on a way to drastically expand the set of organisms and pathways covered by WikiPathways. Yeast has 66 pathways there, compared to 1319 for human. By doing fast ortholog detection at runtime (using another SPARQL API, provided by OrthoDB [4]) I'm hoping to be able to convert relevant annotated pathways across organisms, e.g. human to yeast, mouse to rat, Arabidopsis to rice -- and vice versa.
I created Ideogram.js, a JavaScript library for chromosome visualization [1].
Ideogram supports genomic views to research and report findings on cancer, clinical variants, gene expression, evolution, agriculture, and more [2]. What previously existed for genome visualization was either focused on short genomic regions (e.g. genes) or complex to set up and maintain.
Janssens seems less skeptical of 23andMe's paper on polygenic score for type 2 diabetes [1][2], which -- interestingly -- positively cites the Khera 2018 paper on polygenic score for heart disease that she critiqued. Some researchers are skeptical, but the medical community generally seems to consider polygenic scores promising for tests [3][4].
> you haven't posited how Bob's mapped genome sitting in 23andme will be used for medical treatment.
Early intervention. Polygenic scores could be used for medical treatment by motivating earlier intervention. That could include stronger recommendations for better diet and exercise, closer monitoring programs, or more precise prescriptions. That, in turn, could reduce disease burden.
Large genetic datasets yield medical progress by increasing statistical power of tests. These better tests enable earlier and more targeted treatment.
Take heart disease. It has a significant but complex genetic component. Many genetic variants each contribute a small amount to risk for heart disease. If a given person has many small risk variants, the sum total risk -- often called "polygenic score" -- can be relatively high.
People in the top 8% of polygenic scores had a 3x higher risk for heart disease than the general population [1][2]. Through techniques like polygenic scoring, large genetic datasets enable uniquely early detection of high risk for the world's leading cause of death.
UK Biobank has a smaller cohort but deeper and more reliable data.
All UK Biobank participants were tested for blood pressure, bone mineral density, grip strength, BMI, etc. The last 200,000 participants underwent detailed tests of cognitive function [1].
23andMe asks customers for health information, but self-reports are not usually as reliable as the clinician-administered tests done in UK Biobank.
23andMe is CLIA certified and FDA authorized, and makes many medical claims. From [1]:
"CLIA certification and CAP accreditation
23andMe laboratory testing is done in U.S. laboratories certified to meet CLIA (Clinical Laboratory Improvement Amendments of 1988) standards, including qualifications for individuals performing testing and other standards to ensure the accuracy and reliability of results. The laboratory is also accredited by the College of American Pathologists (CAP), which has served as a model for various federal, state, and private laboratory accreditation programs throughout the world."
FDA authorizations for 23andMe's personal genome service are available online, e.g. [2] for Alzheimer's disease risk reporting based on the E4 variant of the APOE gene.
The company also offers ancestry reports, which are not clinical and thus covered by CLIA. But medical claims in 23andMe's health reports do comply with CLIA and other regulations.
> Manhattan has twice the people with half the land area, on an island.
Manhattan has twice the residents based on U.S. Census population data, but actual daytime population is more than just residents. A more important statistic is commuter-adjusted population [1], i.e. number of people in an area during normal business hours, including workers. That's where Manhattan and San Francisco really diverge.
The commuter-adjusted population of Manhattan is 3.1 M [2], compared to 1.6 M residents [3]. San Francisco has a commuter-adjusted population of 1.0 M [4, 5], compared to 0.8 M residents [5]. In other words, Manhattan's population booms almost 200% during normal business hours, while San Francisco's increases a modest 25%. Manhattan may have 2x more residents, but it has 3x more daytime population than San Francisco.
Given it has about half the land area, Manhattan's daytime population density is thus 6x that of San Francisco.
Attitudes on development -- namely transportation infrastructure and building construction -- certainly contribute to that difference.
[1] https://undark.org/2024/07/09/informed-consent-human-genome-...