Science on Target: Q3 Publication Spotlight

Predicting Phenoconversion in Carriers of ALS Pathogenic Variants

People carrying mutations known to cause ALS  make up at least 10% of the ALS cases.  These mutations are often not fully penetrant, leaving these individuals with uncertainty as to whether they will get the disease, what clinical symptoms will manifest (ALS or FTD), and when this might occur.  Familial ALS (fALS) biomarkers remain elusive and the paucity of biomarker options have led clinicians and researchers to rely heavily on neurofilament light chain (NfL). While NfL is useful to monitor ALS severity and, in some cases therapeutic effects, it is a limited  predictor of disease progression or phenoconversion and can be conflated with neurological injury.  To fill this gap in clinical tools, Ran et al. (Nature Medicine, 2026) recently identified novel biomarkers using state of the art proteomics, machine learning, and plasma from people with pathogenic ALS and ALS/FTD variants who have yet to phenoconvert (pre-symptomatic), those who have phenoconverted, people with ALS, and non-neurological healthy controls. 

Ran et al. began by identifying 137 proteins that were differentially abundant when comparing plasma from people with ALS and non-neurological controls. Subsequent investigation of this first set of proteins was reduced to 73 proteins, following a trajectory analysis of phenoconverters, and then further reduced to 52, which accounted for proteins significantly changing over time with respect to phenoconversion. In an effort to predict the likelihood of phenoconversion, the aforementioned proteins were evaluated using several machine learning methods that compared pre-symptomatic and phenoconverter proteomic profiles over five time-horizons (relative to phenoconversion). 

Unsurprisingly, their data did show that NfL, alone, increases sharply prior to phenoconversion and continues to increase with clinical manifestation and disease progression. Additionally, NfL consistently appeared as the strongest single predictor across all five time-horizons. Using a data-driven approach, the candidate proteins were further refined to a 19-protein and 16-protein panel. The 19- and 16-protein panels outperformed NfL alone in predicting phenoconversion, with the 19-protein panel performing the best with sensitivity and specificity ranges of 76 – 95%. Comparably, NfL alone had sensitivity and specificity ranges of 61 – 81% and 84% – 93%, respectively. Furthermore, the correlation between actual and estimated number of years to phenoconversion was strong and relatively limited mean absolute error. Finally, the 19- and 16-protein panels were evaluated in a replication cohort from a UK Biobank. The replication cohort results suggested that 15- and 10-proteins, from the 19- and 16-protein panels, respectively, continued to outperform NfL in predicting phenoconversion. 

This is the first publication demonstrating the potential for biomarkers to predict when a patient might phenoconvert. This application of machine learning is exciting but warrants caution given that many environmental and genetic risk factors can influence time of disease onset. If replicated in additional cohorts, the protein panels tested in this study have the enormous potential to  improve patient care and  enable clinical trials testing early intervention of novel therapies. 

Related News

Brooke Eby | 1988-2026

October 2, 2026

What are biomarkers in ALS research?

September 16, 2026

TDP-43: What It Is and Why It Matters in ALS

September 16, 2026

Building the Data Foundation for AI-Driven ALS Drug Discovery

August 25, 2026