WEST Algorithm Predicts Rare Diseases Using Electronic Health Records

An artificial intelligence algorithm called WEST, developed by researchers at Harvard Medical School and Boston Children’s Hospital, predicts rare diseases from electronic health records using noisy clinical data.

Rare diseases present a profound public health paradox. While collectively affecting hundreds of millions of people worldwide—including an estimated 30 million in the United States and up to 10 million in Mexico—individual conditions remain isolated across small populations. This fragmentation leaves patients vulnerable to prolonged diagnostic delays, often spending years consulting multiple specialists before receiving a definitive answer.

How the WEST Algorithm Identifies Rare Diseases in Electronic Health Records

To tackle the information gaps that stall diagnoses, researchers created the WEakly Supervised Transformer (WEST) algorithm. Described in research published in npj Digital Medicine, the program uses electronic health record data to predict when a patient may harbor a rare condition. Chan School of Public Health, noted that young patients often exhibit confusing symptom arrays that challenge clinicians who lack prior experience with these conditions.

The algorithm functions similarly to large language models like ChatGPT by taking a patient’s medical history—including diagnostic codes, lab results, and chart notes—and projecting future diagnostic sequences. Its core innovation lies in weak supervision, allowing the model to learn from patients both with and without confirmed diagnoses. Tianxi Cai, a professor at Harvard Medical School, explained that the team treats imperfect data as labeled to guide the model while acknowledging its flaws.

Scaling Whole-Genome Sequencing and Long-Read Diagnostics

Beyond electronic health records, diagnostic laboratories are turning to advanced genomic technologies to bypass the limitations of traditional testing. Standard exome sequencing reads only protein-coding regions of DNA, missing regulatory noncoding segments. To capture the full genomic picture, researchers explore whole-genome sequencing and long-read methods.

Alexander Hoischen, a genomic technologies researcher at Radboud University, utilizes long-read sequencing to read fragments up to 20,000 base pairs, overcoming the puzzle-like gaps of short-read sequencing.

Institutional partnerships are also expanding clinical access. Illumina announced that its Laboratory Services unit is providing clinical genome sequencing and interpretation to the diagnostic lab at the Florida Institute for Pediatric Rare Diseases (IPRD) at Florida State University.

Eric Green, chief medical officer of Illumina, noted that clinical whole-genome testing provides a transformative opportunity to end the diagnostic odyssey for the estimated 30 million people in the United States living with a rare disease.

Grants, Pangenomics, and Global Research Initiatives

Efforts to accelerate discovery are receiving targeted financial and technological backing. Anthropic launched a focused call for applications for its AI for Science program centered on rare genetic diseases, offering accepted applicants up to $50,000 in Claude credits over six months. The grant program includes a track for basic science partnerships and a second track for early-stage biotechs accelerating clinical development.

Hand with protective shield and network node in cybersecurity symbol design — Photo: Anthropic

In basic science, grantees are encouraged to utilize resources like the Monarch Initiative’s DisMech library, where AI tools can ingest case reports and variant databases to uncover mechanistic similarities across distinct conditions.

Structural and Regional Barriers to Timely Care

Despite technological leaps, real-world delivery remains starkly uneven. In Mexico, where up to 10 million people are affected by rare diseases, patient advocates and industry leaders point to an average diagnostic delay of seven to eight years. David López, managing director for BioMarin Pharmaceuticals México, noted that late diagnoses allow the silent progression of disease to impair vital functions before intervention occurs.

WEST Algorithm Predicts Rare Diseases Using Electronic Health Records — Photo: The Scientist

Stakeholders emphasize that translating technological innovations into equitable clinical care requires structural reforms in regulatory timelines, financing, and medical education to ensure primary care physicians recognize early symptoms before patients cycle through years of uncoordinated specialist referrals.