AI cracks the multiple sclerosis code

Overview

Scientists at UCL used AI to identify two biologically distinct subtypes of multiple sclerosis (early sNfL and late sNfL) based on when nerve damage shows up in blood tests and brain scans. The discovery, published in Nature Medicine in December 2025, means doctors can finally match 2.9 million MS patients worldwide to treatments that actually fit their disease progression pattern.

This isn't incremental progress. For decades, MS treatment has been one-size-fits-all: wait for symptoms, prescribe drugs, hope they work. The SuStaIn model analyzed data from 600 patients, finding two patterns: early aggressive corpus callosum lesions, and slow brain shrinkage in the limbic cortex.

Patients with early sNfL could now get high-efficacy drugs immediately instead of after they've already lost function. The stakes are whether MS care joins the precision medicine revolution—or stays stuck in trial-and-error mode.

Play on this story Voices Debate Predict

Key Indicators

New MS subtypes identified

Early sNfL and late sNfL subtypes based on timing of nerve damage biomarkers

2.9M

People with MS worldwide

Global patient population who could benefit from personalized treatment

634

Patients in UCL study

Dataset size for AI model training combining blood tests and brain scans

144%

Increased lesion risk

Early-sNfL patients show 144% higher risk of new brain lesions vs late-sNfL

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Play

Exploring all sides of a story is often best achieved with Play.

Predict 3 ways this could play out. Contrarian picks score more — points lock when the scenario resolves. Log in to play

Connections Sixteen names from the news. Find the four hidden groups of four. Log in to play

People Involved

Dr. Arman Eshaghi

Lead researcher developing MS-PINPOINT AI tools for personalized MS treatment

Prof. Frederik Barkhof

Senior advisor on imaging biomarkers for MS research

Dr. Alexandra Young

Co-developer of SuStaIn algorithm

Dr. Neil Oxtoby

Leading clinical translation of SuStaIn algorithm

Organizations Involved

Queen Square Analytics

UCL Spinout Company

Commercializing AI algorithms for neurological disease analysis

UCL spinout developing breakthrough imaging analysis technology for clinical trials and patient care in neurological diseases.

UCL Queen Square Institute of Neurology

Academic Research Institute

World-leading neurology research center

Premier neurological research institute developing AI tools for understanding brain disease progression and treatment.

Timeline

October 2018 January 2026

7 events Latest: January 5th, 2026 · 5 months ago

Tap a bar to jump to that date

January 2026
UCL Announces MS Subtype Discovery
Latest Research

UCL publicly announces Brain journal findings, triggering international media coverage and discussion in MS patient communities about implications for personalized treatment.
January 5th, 2026
December 2025
AI Identifies Two New Biological Subtypes of Multiple Sclerosis
Research

UCL and Queen Square Analytics publish Nature Medicine study identifying early sNfL and late sNfL subtypes using blood biomarker analysis combined with brain imaging, enabling personalized treatment strategies.
December 30th, 2025
Research Published in Brain Journal
Research

UCL and Queen Square Analytics publish study in Brain (Volume 148, Issue 12) identifying early-sNfL and late-sNfL subtypes, with early-sNfL patients showing 144% increased risk of new lesion formation.
December 4th, 2025
November 2022
MS-PINPOINT Project Funded
Funding

Arman Eshaghi receives NIHR Advanced Fellowship to develop AI tools predicting individual MS patient outcomes and treatment responses.
November 1st, 2022
April 2021
First AI-Identified MS Subtypes Discovered
Research

UCL team uses SuStaIn to identify three MS subtypes from 6,322 MRI brain scans, demonstrating AI's potential for disease classification.
April 1st, 2021
January 2020
Queen Square Analytics Founded
Commercial

UCL spinout launched to commercialize SuStaIn algorithm and bring AI-driven neurological disease analysis to clinical practice.
January 1st, 2020
October 2018
SuStaIn Algorithm Debuts in Nature Communications
Research

UCL researchers publish machine learning technique that identifies disease subtypes and progression stages from cross-sectional data, initially applied to Alzheimer's disease.
October 1st, 2018

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

2020-2024

AlphaFold Cracks Protein Structure Problem

DeepMind's AlphaFold 2 solved the 50-year protein structure prediction challenge in 2020, achieving 90%+ accuracy on protein shapes. By 2022, AlphaFold predicted structures for 200 million proteins—a thousand-fold increase over experimental data. Over 3 million researchers in 190 countries adopted the tool for drug discovery, vaccine development, and understanding disease mechanisms.

Then

Academic labs and pharma companies immediately integrated AlphaFold into research pipelines, accelerating drug discovery timelines by months to years.

Now

Established AI as credible for solving fundamental biological problems, attracting billions in biotech AI investment and enabling the current wave of disease-subtyping algorithms.

Why this matters now

AlphaFold proved AI could handle biological complexity that defeated traditional methods. The MS subtyping breakthrough follows the same playbook: using pattern recognition on complex data to reveal hidden structure that changes how we approach treatment.

2011-2022

IBM Watson for Oncology Failure

After winning Jeopardy in 2011, IBM positioned Watson as cancer care's future. But Watson for Oncology was trained on hypothetical cases from one hospital, not real patient data. It gave unsafe treatment recommendations, performed worse than human oncologists in trials, and failed to demonstrate patient benefit despite partnerships with 50+ hospitals. IBM sold off Watson Health assets in 2022 after a $4 billion loss.

Then

Hospitals abandoned Watson implementations; oncologists lost trust in AI diagnostic claims; investors became skeptical of overhyped medical AI.

Now

The failure taught medical AI developers that training data quality, clinical validation, and narrow problem definition matter more than general intelligence or marketing hype.

Why this matters now

The UCL approach learned Watson's lessons. SuStaIn uses real patient data from 600+ MS cases, focuses narrowly on subtype identification rather than treatment recommendation, and publishes peer-reviewed validation before commercializing. Success requires rigorous science, not just powerful algorithms.

1993-2010

First MS Disease-Modifying Therapy Approved

In 1993, interferon beta-1b became the first drug proven to alter MS's natural course, reducing relapses in relapsing-remitting MS by about 30%. This was revolutionary after a century of MS being untreatable. By 2010, several more disease-modifying therapies emerged (glatiramer acetate, natalizumab, fingolimod), but all followed an escalation approach: start with moderate drugs, switch to high-efficacy therapies only after disease worsens.

Then

MS transformed from progressive disability sentence to manageable chronic condition for many patients; neurologists gained treatment options beyond symptom management.

Now

The escalation paradigm became entrenched despite evidence that early aggressive treatment prevents irreversible damage better than waiting for progression.

Why this matters now

For 30+ years, MS treatment has been one-size-fits-all escalation. The AI subtyping breakthrough challenges this orthodoxy by enabling personalized treatment from day one—early sNfL patients get high-efficacy drugs immediately, potentially preventing damage the escalation approach allows to accumulate.