Learn genomics by travelling through it

a very long book written in a four-letter alphabet, copied into every room of a house.

Everything BharatGenome compares is ultimately a difference in this text between groups of people.

an alphabet with four letters instead of twenty-six. Meaning comes from the order, not the number of letters.

When we say a population carries 'A' or 'G' at a position, these are the letters we mean.

chapters of the book, each bound separately so the whole thing stays organised.

A variant is located by chromosome and position — that coordinate is how two datasets know they are talking about the same spot.

one recipe inside the cookbook. The whole book is not a recipe; a specific passage is.

Searching a gene symbol in Explore is usually how a researcher starts, because genes are the unit most research is organised around.

the entire cookbook, not just one recipe.

'Genomic representation' is about whose whole genomes were read to build the datasets everyone else relies on.

a typo that appears in some printings of a book but not others. Most typos change nothing; a few change the meaning.

Variants are the objects BharatGenome compares. How common a variant is can differ a lot between populations.

a single-character typo, as opposed to a whole missing paragraph.

Most frequency comparisons — and most of the research that depends on them — are about SNPs.

the two spellings of a word that both appear in print — 'colour' and 'color'. Each spelling is an allele.

The Indian-vs-global comparison is literally a comparison of how common each allele is.

a survey sample. 'What do people think?' is meaningless until you say which people were asked.

India is not one population — it is many, with deep structure. Even 'Indian' is a simplification, and BharatGenome says so on every result.

asking a crowd which spelling they use and reporting the split. The answer describes that crowd — ask a different crowd and you may get a different split.

This is the single number at the heart of BharatGenome. The whole product exists because it is not the same everywhere.

an ISBN for a book. Different shops, different catalogues, same number.

It is the cleanest way to ask two different datasets about exactly the same variant — which is what a comparison requires.

street numbering for a city. Useful because everyone agrees on it, not because house number 1 is the best house.

Coordinates come from the reference. It is a standard for locating things, and should not be read as what a genome 'should' look like.

a phrasebook compiled from certain speakers. It works well for dialects it covers and poorly for ones it does not.

If a panel contains few Indian genomes, the results it produces for Indian samples are weaker — a direct, practical consequence of representation.

autocomplete. It guesses the rest of the word from the letters typed — and it guesses much better in a language it was trained on.

Imputation quality depends on the panel matching the person. Under-represented populations get less accurate imputation.

words that usually travel together in a phrase. Hear the first and you can often guess the second — but which words pair up differs by dialect.

LD patterns differ between populations, so a result anchored to one population's LD structure may not point at the same thing in another.

comparing two large groups' shopping lists to spot items that show up far more often in one.

GWAS results reflect the cohort that was studied. If Indian cohorts are absent, the associations found may not be the ones that matter here.

a credit score built from many small signals — and calibrated on one particular set of customers.

PRS are known to lose accuracy when applied to populations unlike the one they were derived from. That loss is the representation problem in its most visible form.

different people metabolising caffeine at different rates. Same cup, different effect.

The frequency of relevant variants differs between populations, so population-specific frequency data has direct research value here. BharatGenome reports frequencies only; it gives no clinical or dosing guidance.

a national survey that mostly polled one region. The findings are real, but they describe that region best.

This is the core problem statement. India's share of global genomic reference data is far smaller than its share of human genetic diversity.

a recipe developed at sea level. It still works in the mountains, but not without adjustment.

BharatGenome's comparison is a transferability check: does the global number describe the Indian population, or not?

a spell-checker trained on one dialect. It is not broken — it is simply calibrated for someone else.

It explains why this is a methodological issue with measurable consequences, not a matter of fairness alone.

a library with one wall of shelves left largely unfilled — you cannot tell what you are missing until you look.

It is the reason BharatGenome makes Indian evidence the default side of every comparison rather than an optional filter.

a second opinion printed next to the first, with both doctors' credentials attached.

Making the gap visible and quantified is what turns a known problem into something a researcher can act on.