Artificial intelligence made concrete advances in medicine during 2025, with new models improving drug discovery, diagnostics and clinical workflows. Several open‑source systems sped research pipelines, while Indian institutes deployed practical clinical tools to support frontline care. These developments look set to accelerate treatments for rare diseases, improve diagnostic equity and reduce clinician workload.
AI in healthcare 2025
One of the year’s most notable advances came from collaboration between academic and industry teams. Boltz‑2, developed by MIT and Recursion Pharmaceuticals, offers ultra‑fast prediction of protein–ligand structures and binding affinity, enabling much quicker virtual screening. Released as open source, the model has been used to compress early drug discovery timelines by prioritising likely candidates for laboratory testing.
In genomics, Evo 2 emerged as a broad foundation model trained on over 128,000 genomes from more than one lakh species. The model can generate and predict novel DNA, RNA and protein sequences, expanding possibilities for designing biological constructs and engineered treatments for rare genetic conditions.
Quality and equity in diagnostics were also priorities. Harvard Medical School published a pathology bias correction tool after detecting bias in earlier diagnostic models. This new approach reduces population disparities in cancer detection, improving diagnostic performance across diverse cohorts.
India contributed several high‑impact models and deployments. OncoMark, developed by the S. N. Bose National Centre for Basic Sciences with Ashoka University, is an open‑source model trained on 3.1 million single cells spanning 14 cancer types. It predicts interactions among hallmarks such as metastasis and immune evasion, informing personalised therapy strategies and research into therapy resistance.
At the clinical front line, AIIMS New Delhi unveiled Smart Doctor, a clinical decision support system intended for rollout across some 70,000 public and private hospitals nationwide. Smart Doctor aims to reduce medical errors and standardise diagnostic and treatment pathways for long‑term and non‑communicable diseases, supporting clinicians rather than replacing them.
Screening programmes received particular attention. MadhuNETrAI, a collaboration between AIIMS, the Union Health Ministry’s e‑health division and Wadhwani AI, enables non‑specialist health workers to screen for diabetic retinopathy using AI tools. The system went live across 38 centres in 11 states in late 2025, promising earlier detection and referral for vision‑threatening disease.
Maternal health saw improvements through Garbhini‑GA2, an India‑specific model developed by BRIC‑THSTI and IIT Madras to improve fetal age estimation. Clinical validation carried on through 2025 after the model’s 2024 unveiling. Western models can misestimate gestational age in Indian pregnancies by a week or more; Garbhini‑GA2 reduces that error to under half a day, which is critical for preterm birth prediction and timing of interventions.
Taken together, these models reflect a trend towards open science, local validation and clinical integration. Researchers and health systems emphasised transparency, bias correction and real‑world testing as essential steps before broad deployment. Regulators and clinical leaders will need to ensure proper oversight, data governance and training for staff using these tools.
As 2025 closed, AI systems were shifting from experimental proofs of concept to tools that can be audited, validated and scaled. The combination of global research advances and India’s practical deployments suggests faster translation from model development to patient‑facing care, with potential benefits for health outcomes and equity.
Key Takeaways:
- Major AI models in 2025 accelerated drug discovery, diagnostics and genomics, marking progress in AI in healthcare 2025.
- Open‑source platforms such as Boltz‑2 and Evo 2 sped virtual screening and sequence design, shortening research timelines.
- Indian institutions delivered clinical tools—OncoMark, Smart Doctor, MadhuNETrAI and Garbhini‑GA2—targeting cancer care, diabetic retinopathy screening and accurate gestational dating.
