MISHA

An independent AI lab for rare disease

One rare disease is rare. All rare diseases are not.

It started with my son's genetic hearing loss. Now it is a lab for the rare diseases the market skips.

Egor Lyfar teaching an AI for Scientific Research masterclass at the University of Hong Kong
Me at the University of Hong Kong, ranked first in Asia, teaching my AI for Scientific Research masterclass. The method this foundation runs on.

Rare, one at a time. Common, all at once.

Any single rare disease is, by definition, small. Put them together and they are one of the largest health problems on earth. That is the whole thesis. The hard part is not any one disease. It is that the work does not compound across them. Registries do not talk. Tools get rebuilt. Each family starts from zero.

7,000+ distinct rare diseases and the count rises every year as sequencing finds new causes NIH GARD
~300M people affected worldwide a conservative estimate of 263 to 446 million Nguengang Wakap et al., 2020
72% are genetic in origin mostly single-gene, which is exactly what computation can model Nguengang Wakap et al., 2020
<5% have an approved treatment so more than 95 in 100 diagnoses come with nothing to offer NIH GARD
Sources (3)

How MISHA works

The bridge, and the guard on the bridge.

Three groups need each other and cannot connect. Families live the disease but cannot turn it into research. Researchers can test ideas but are scarce and flooded with noise. Patrons can fund it but need to see leverage. MISHA is the connective tissue between them, and the filter that keeps it honest.

How we work

The real product is a research pipeline. Hearing loss is where we point it first.

We test one AI-augmented research pipeline on open problems in pure mathematics, where a proof kernel checks every step, and point the same pipeline at rare genetic disease. The method is the product. The disease is the proof.

How the pipeline works

This one started at a hearing test.

My son Misha was diagnosed at age two with bilateral sensorineural hearing loss caused by mutations in the STRC gene, the condition known as DFNB16. STRC is the second most common genetic cause of this kind of hearing loss. It is common enough to matter and rare enough for the drug industry to skip. That gap, between a solvable problem and a funded one, is where we work.

#2 most common genetic cause of recessive nonsyndromic hearing loss, after GJB2 Frontiers in Genetics, 2021
0 approved therapies for STRC / DFNB16 hearing loss today MISHA vault
0 labs in Asia working on it, while a handful work elsewhere MISHA vault
2 years old at diagnosis the cochlear development window does not wait for funding family

There is proof this is solvable. In 2026, a dual-AAV gene therapy restored hearing to near-normal thresholds in a DFNB16 mouse model, and the animals regained normal frequency discrimination for up to 100 days. The biology is tractable. What is missing is coordination and money, not possibility.

Read the mission