A paper published in npj Digital Medicine demonstrates how the ALS Therapy Development Institute (ALS TDI) and Google harnessed patient data and machine learning to develop objective measures of ALS symptom severity.
Read the summary below and attend our upcoming Town Hall to hear Dr. Vieira discuss these new tools and see a live demonstration!
Because ALS clinical presentation varies so much from person to person, it has been extremely difficult to measure whether or for whom treatments for ALS are effective.
Having better tools to assess ALS disease severity and progression could have a huge impact on the development and delivery of effective treatments to people with ALS. For this reason, a major goal of ALS TDI's Precision Medicine Program (PMP) has been to discover new biomarkers, or objective measures of ALS.
In a recently published paper, ALS TDI and Google authors describe how they used PMP data and machine learning (ML), to develop digital biomarkers of ALS - new tools that can objectively measure ALS symptom severity and progression over time.
“This paper, and the tools that it describes, are the culmination of years of work. It started with conception of ALS TDI’s Precision Medicine Program in 2014, complete with collection accelerometer data and voice recordings - long before most people were thinking about how to use machine learning and A.I. to study and measure disease symptoms,” said co-lead author and ALSTDI CEO/CSO, Fernando G. Vieira, M.D.. “But more than anything, it’s the culmination of a herculean effort by our patient participants who contributed the necessary data around which we could collaborate with the great minds at Google to take these necessary steps in ALS biomarker development.”
The Critical Role of PMP Participant Data
The development of these tools was only made possible thanks to years of patient data gathered through ALS TDI's Precision Medicine Program (PMP), as well as the support of the ALS community.
Thanks to ALS TDI's partnership with Google, these years of research have now resulted in the creation of algorithms that use digital outcomes that can predict a person’s ALSFRS-r score in the movement and speech categories with more than 75% accuracy. Where the algorithm and a person’s ALSFRS-r score disagree may be indications that the algorithms are actually performing better than ALSFRS-r.
“ALS TDI has collected an incredibly large dataset of accelerometer and voice samples among other data from people living with ALS, remarkable in terms of number of patients as well as duration of collection,” said co-lead author Subhashini Venugopalan of Google, “This type of project might be a template for other efforts in precision medicine and drug development across other disease indications.”
Predicted scores generated by the voice algorithm are now available to participants in the PMP through their personalized online portal.
"The ALS-FRSr has always frustrated me since it's so subjective and broad," said Andrea Peet, a PMP participant. "My speech has been 'intelligible with repeating" for at least four years now, but I know it has gotten worse. Having a tool like this algorithm to more precisely and objectively measure progression makes total sense. I hope I live to see the day when this tool can be used to tell if a clinical trial drug is actually working."
Teaching Computers to Measure ALS Symptoms
To create these tools, Google researchers utilized machine learning to develop an algorithm that can analyze PMP voice recordings and movement data and assign ‘scores’ based on a person’s performance.
These powerful ML tools could be a step toward allowing clinical investigators to more accurately measure changes in symptom progression and learn how they relate to experimental ALS treatments and interventions in clinical trials.
In addition to improving interpretability of clinical trials, these tools might also help patients and their doctors better understand their own disease progression - enabling them to create more effective treatment plans tailored to each individual patient's needs.
How These Tools Can Impact ALS Research
These tools developed by ALS TDI and Google are now freely available to clinical investigators and other researchers. The code for the algorithms have been posted on GitHub. This open access code repository will allow anyone to access it and, potentially, even make changes to improve upon it.
The goal is, ultimately, to catalyze the development of even better digital measures for ALS progression that can be used in clinical drug development and patient care.
In recent years, other groups led by investigators including Drs. James Berry of Massachusetts General Hospital and Jeremy Shefner of the Barrow Neurological Institute, as well as non-profit groups like AnswerALS and EverythingALS, have contributed to large efforts to gather data from people with ALS and learn more about the disease. These efforts each introduce new and innovative techniques - building on previous findings.
"As the Drug Discovery Engine for ALS, ALS TDI’s ultimate goal is to continue to invent and discover drugs until everyone with ALS has effective treatments. By developing tools to more accurately measure progression we hope to be able to better match the right treatments to the right subsets or ALS." – Fernando Vieira M.D.
How to Contribute
People with ALS who are interested in accessing their data and advancing ALS research can enroll in the PMP for free by visiting www.als.net/precision-medicine.
If you want to talk to someone at ALS TDI to learn more about our research, you can contact us here.
What to do next
• Read the paper, A Machine-Learning Based Objective Measure for ALS Disease Severity.
• Attend our upcoming Town Hall to learn more about these digital biomarkers.