The Hyderabad-based International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) will work with Google DeepMind, Google.org and two US genomics groups to build an open-access pangenome atlas of sorghum. The project will apply machine learning to genetic material from ICRISAT’s genebank. DeepMind is also funding its first postdoctoral fellowship outside the UK at ICRISAT, to develop genomic tools for chickpea, groundnut and pigeon pea.
As Telangana confronts one of its worst droughts in years, a partnership announced on Thursday is aimed at the crop that may matter most in a hotter, drier future. On 15 October, Google described a collaboration between Google DeepMind, Google.org’s AI Collaborative for Food Security, ICRISAT, the HudsonAlpha Institute for Biotechnology in Alabama and the University of Washington’s Department of Genome Sciences to build an open-access genetic atlas of sorghum, known in India as jowar.
The atlas will draw on a representative sample of ICRISAT’s genebank, which, according to Google and ICRISAT, holds more than 44,000 genetic samples from over 90 countries, including India. Machine learning will be combined with genetic analysis to identify promising combinations of traits, so that breeders can develop resilient varieties faster than through conventional trial-and-error breeding, which the partners say can take decades.
Why sorghum
Sorghum is one of the world’s most important cereals for dryland farming. It is a C4 plant, using a form of photosynthesis that is efficient in heat and strong light, and it tolerates drought and poor soils better than rice, wheat or maize. Its grain feeds people, its stalks feed livestock, and sweet-sorghum varieties are used for biofuel. ICRISAT says sorghum sustains the nutrition of more than 500 million of the world’s most climate-vulnerable people, mostly in Africa and Asia.
In India, sorghum belongs to the millet family that the government promotes as Shree Anna, and it was central to the UN’s International Year of Millets in 2023. Yet the area under jowar in India has declined over decades as irrigation and procurement policies favoured rice and wheat. The current drought, in which farmers are being asked to defer water-hungry rabi paddy, is a reminder of why that balance is being reconsidered.
What a pangenome adds
A reference genome is a single representative sequence of a species’ DNA. Sorghum’s first reference genome, from the inbred line BTx623, was published in 2009. A single reference, however, misses much of a crop’s diversity. Genes present in a drought-adapted landrace from Ethiopia or a fodder variety from Maharashtra may simply be absent from the reference line.
A pangenome assembles the genomes of many diverse varieties, capturing genes and structural variations that exist only in some of them. For breeders, these “accessory” genes are often where the useful traits lie, such as tolerance to heat, drought, pests or poor soils. Building a pangenome from a genebank’s diversity turns stored seed into searchable genetic information.
The role of machine learning is to make that information usable. Linking millions of genetic variants to traits measured in the field is a large statistical problem. Models can help predict which genes or combinations of variants are likely to improve a trait and which crosses are worth making, narrowing the options before seeds go into the ground.
The people and the plan
ICRISAT’s Damaris A. Odeny, the project’s principal investigator, said that “pangenomes are valuable resources for accelerating the development of climate-resilient crop varieties”. Stanford Blade, ICRISAT’s Deputy Director General for Research and Innovation, said: “We are excited to partner with Google and bring AI’s potential to the mission of transforming dryland agriculture.” Abi Ramanan, who leads Google DeepMind’s food-security portfolio, said that by open-sourcing what ICRISAT calls “the first Global Dryland Sorghum Pangenome”, the partners aim to empower breeders and scientists.
The second element is a three-year Google DeepMind postdoctoral fellowship at ICRISAT. The fellow will build tools to characterise the genomes of chickpea, groundnut and pigeon pea (arhar or tuar dal). These are staple pulses and oilseeds for India’s drylands, and all three are crops for which ICRISAT has a global breeding mandate. Google says it is DeepMind’s first fellowship in India and its first outside the UK.
The announcement was part of a wider Google post on its AI work in India. Google’s Agricultural Understanding models are now in public preview for developers. Through the Gujarat government’s Krushi Pragati platform, built by Amnex, they have supported more than 900 million SMS advisories to over 600,000 farmers across more than 14 million hectares in the past year, according to Google.
ICRISAT brings more than seed to the partnership. Founded in 1972, with its headquarters at Patancheru near Hyderabad, the institute has bred and released sorghum, pearl millet, chickpea, pigeon pea and groundnut varieties across Asia and Africa for five decades, often through national partners such as ICAR. Its genebank is one of the world’s largest collections of dryland crops, and its breeders hold decades of field data from trials in hot, rain-fed environments. Those trait records are what turn a pangenome from a catalogue of DNA into a breeding tool.
What it is not, yet
This is a partnership announcement, not a scientific result. No funding amount, timeline, sample size or specific AI model has been disclosed, and no peer-reviewed findings have been published. Pangenome projects are demanding: sequencing and assembling dozens or hundreds of genomes, standardising field data on traits, and ensuring that open data does not disadvantage the countries and farming communities that originally conserved the seeds. Access and benefit-sharing for material held in international genebanks are governed by the International Treaty on Plant Genetic Resources for Food and Agriculture, and how the atlas handles this will be closely watched.
Faster genetic discovery also does not automatically produce better seed in farmers’ hands. New varieties still need field trials, release through national systems and seed multiplication, and markets and procurement must make sorghum worth growing.
Why it matters for Hyderabad
ICRISAT, at Patancheru on Hyderabad’s outskirts, has long been a quiet asset of the region’s science. Linking its genebank to one of the world’s leading AI laboratories places Hyderabad at the centre of a global effort on climate-resilient crops, at a moment when the state’s own farmers are living through the consequences of climate stress.
