Due to its remarkable variability, Multiple Sclerosis has long challenged scientists and clinicians—but with the rise of in silico medicine, new breakthroughs are finally on the horizon.
Multiple Sclerosis (MS) is often called the disease of a thousand faces. And for good reasons– no two patients experience it the same way. One day, vision blurs. Another day, legs feel heavy. Then come months or even years of near normalcy, only for symptoms to return like unwelcome guests. It’s unpredictable, relentless, and unique in the way patients experience it.
Multiple sclerosis is the most common cause of neurological disability in young adults, a chronic disease where the immune system mistakenly attacks the protective sheath–myelin– around nerve fibres in the brain and spinal cord. Depending on which area is affected, the symptoms vary: vision loss, speech difficulties, fatigue, mobility issues, and even severe disability. Sometimes the disease progresses steadily; other times, it ebbs and flows in alternating periods of worsening and recovery, referred to as relapsing-remitting rhythm.
Currently available treatments only allow for the slowdown of the disease progression. Moreover, the above variabilities make the diagnosis and prognosis of MS challenging. In the absence of a single definitive test for MS, clinicians must rely on a combination of clinical signs, MRI scans, and spinal fluid analysis to reach a diagnosis.
Roughly 80-90% of patients are initially diagnosed with the relapsing-remitting form. Over time, many of these patients–if untreated– will progress to a secondary progressive phase, where neurological decline continues progressively, without any remission.
Identifying who will progress–and when– is crucial for timely, effective treatment. But right now, there’s no clinically validated way to predict the progression of the disease within the first five years of diagnosis. And this is one of the challenges tackled by researchers using in silico methodologies.
Since 2000, scientists and doctors at Brigham and Women’s Hospital in Boston have been enrolling MS patients in the CLIMB study. By 2017, they collected enough data to try something new: use machine learning to predict disease progression.
In a study published in PLOS ONE, researchers trained a machine learning model using data from 1,352 patients, including 525 who developed the progressive form of MS within five years. Initially, the model was tested using only baseline data. While its performance exceeded random guessing, it was not clinically reliable, as it disproportionately classified patients as non-progressive—the majority group. However, when data from the one-year follow-up and MRI scans were included, the model’s accuracy in predicting progression jumped to over 80%.
In 2020, a second wave of research built on earlier efforts by combining data from the CLIMB study with the EPIC dataset from the University of California, San Francisco, where patients have been followed since 2004. Researchers tested six machine learning models, including advanced ensemble learning methods—techniques that integrate insights from multiple machine learning classifiers to improve performance. These ensemble models stood out for their consistency and robustness across both datasets, marking a significant step forward in predicting the course of multiple sclerosis.
Meanwhile, in Italy, a valuable contribution has been developed at the University of Catania, which is also a supporting member of the VPH, represented by Prof. Francesco Pappalardo, who, together with his team, developed the Universal Immune System Simulator (UISS). UISS is a powerful digital model of the human immune system capable of simulating various conditions, including multiple sclerosis.
UISS uses a multi-layered architecture that includes physiology, pathology, and treatment layers. The physiology layer models how the immune system functions under normal conditions; the pathology layer simulates how MS disrupts these processes; and the treatment layer predicts how different therapies might affect disease progression. This allows for simulations at both the population level and tailored, individual patient scenarios.
Prof. Pappalardo’s group has published numerous studies using UISS-MS, exploring a wide range of applications—from diagnosis and disease progression to treatment simulations, in silico clinical trials, and beyond, further pushing the frontiers of what we know about multiple sclerosis.
However, despite the promise of these and other innovative tools, significant challenges remain. A key issue is the limited availability of large, high-quality datasets. Multiple sclerosis research still lacks a standardised protocol for data collection and evaluation, making it difficult to compare and integrate findings. Additionally, datasets must be sufficiently large and diverse to avoid biases—particularly those stemming from the underrepresentation of certain racial or ethnic groups. On top of that, data privacy and ethical concerns add further complexity, underscoring the need for a multidisciplinary approach.
But the future is clear: in silico medicine will offer sharper tools to diagnose MS earlier, predict its course more accurately, and test treatments virtually before trying them on real patients.
In the battle against this shape-shifting disease, algorithms aren’t replacing doctors–but they are becoming powerful allies.
Further resources:
Exploration of machine learning techniques in predicting multiple sclerosis disease course
Ensemble learning predicts multiple sclerosis disease course in the SUMMIT study
ICPR 2024 Competition on Multiple Sclerosis. Lesion Segmentation - Methods and Results
Enhancing Multiple Sclerosis Lesion Segmentation in Multimodal MRI Scans with Diffusion Models
Boosting multiple sclerosis lesion segmentation through attention mechanism
In silico clinical trials for relapsing-remitting multiple sclerosis with MS TreatSim
Computational models to predict disease course and treatment response in multiple sclerosis
