Singapore's AI for Science Initiative, a S$120 million programme led by the National Research Foundation, launched eight projects in June 2026, four of them hosted at NUS. The organising idea is less about applying existing models than about pairing artificial intelligence researchers with domain experts to produce what the programme calls bilingual scientists, fluent both in machine learning and in a substantive scientific field such as life sciences or materials.
Two of the NUS projects target physical science and software. Materials Data Foundry, co-led from the NUS Institute for Functional Intelligent Materials with the University of Toronto, uses a combined artificial intelligence and robotics laboratory to assemble a large dataset linking how a material is synthesised to how it performs, with industrial partners including Nvidia and VeChain. AI for Program Reasoning, led from the NUS School of Computing with Imperial College London, builds tools to verify the correctness and security of code as machine-generated software proliferates.
The other two address biology and food systems. A genomics project led from the Yong Loo Lin School of Medicine with A*STAR is developing MultiOmicsFM, a foundation model that interprets DNA, RNA and gene activity jointly for disease-risk prediction, drawing on Singapore's multi-ethnic genomic resources. KGAI4Ag, led from the Department of Biological Sciences with Illinois ARCS, builds agricultural digital twins aimed at climate-resilient farming in Southeast Asia.
What distinguishes the portfolio is its insistence on data generation rather than data reuse. Two of the four projects are explicitly building the datasets their models will need, whether through automated synthesis laboratories or through population genomics. For a field in which progress is increasingly limited by the availability of well-structured scientific data, that emphasis is arguably more consequential than the model architectures themselves.