BEGIN:VCALENDAR VERSION:2.0 PRODID:-//linuxsoftware.nz//NONSGML Joyous v1.4//EN BEGIN:VEVENT SUMMARY:Temporal Phenotyping of ALS Patients using Machine Learning DTSTART:20250704T120000Z DTEND:20250704T140000Z DTSTAMP:20260627T000838Z UID:6d7767b8-1aab-409e-8a9d-8b269c504b3d SEQUENCE:2 CREATED:20250702T075555Z DESCRIPTION:Password: 561232Amyotrophic Lateral Sclerosis (ALS) is a fatal neurodegenerative disease characterized by highly heterogeneous and rapid ly progressing motor decline. Despite extensive research\, predicting ALS progression remains a major clinical challenge\, limiting timely intervent ions such as Non-Invasive Ventilation (NIV) and Percutaneous Endoscopic Ga strostomy (PEG). This thesis addresses this challenge by applying machine learning (ML)-based temporal phenotyping to characterize evolving patient profiles from static and longitudinal clinical data\, with the objective o f uncovering distinct progression trajectories and improving prognosis pre diction.Specifically\, the T-Phenotype algorithm is adapted to the Lisbon ALS Clinic dataset\, leveraging phenotypic predictive clustering to group patients by both outcome and disease trajectory similarities. Comprehensiv e preprocessing strategies were developed to handle irregular clinical eve nt timing\, missing data\, variable-length time series\, and class imbalan ce. The algorithm identified clinically meaningful phenotypes associated w ith the need for NIV (endpoint C1) and PEG (endpoint C3)\, with optimal pe rformance in shorter observation windows (6–12 months)\, revealing disti nct risk profiles aligned with respiratory and bulbar function decline. Co mbining endpoints into a multiclass task (C1+C3) highlighted challenges wi th class imbalance and reduced prediction performance\, although still pro duced interpretable phenotypes. Temporal analysis of cluster transitions i llustrated the model’s ability to capture diverse ALS progression patter ns dynamically\, emphasizing the disease'\;s heterogeneity and the lim itations of static classifications. Although the T-Phenotype model had sat isfactory results in predictive accuracy and clustering interpretability f or the binary endpoints (Hprc metric around 80%)\, its performance decline d for the combined endpoint (Hprc metric around 65%)\, its performance dec lined for the combined endpoint. Additionally\, it lacked sensitivity to c linical improvements post-intervention\, underscoring the need for richer datasets and methodological rigor.The potential of ML-based temporal pheno typing was demonstrated as a valuable tool for understanding ALS progressi on and supporting personalized prognosis and intervention planning. By int egrating temporal dynamics into patient stratification\, this approach adv ances data-driven ALS monitoring and highlights critical considerations fo r future research in this domain. LAST-MODIFIED:20250702T075611Z LOCATION:Online URL:http://df.vps.tecnico.ulisboa.pt/pt/eventos/temporal-phenotyping-of-al s-patients-using-machine-learning/ X-ALT-DESC;FMTTYPE=text/html:

Password: 561232

Amyotrophic Later al Sclerosis (ALS) is a fatal neurodegenerative disease characterized by h ighly heterogeneous and rapidly progressing motor decline. Despite extensi ve research\, predicting ALS progression remains a major clinical challeng e\, limiting timely interventions such as Non-Invasive Ventilation (NIV) a nd Percutaneous Endoscopic Gastrostomy (PEG). This thesis addresses this c hallenge by applying machine learning (ML)-based temporal phenotyping to c haracterize evolving patient profiles from static and longitudinal clinica l data\, with the objective of uncovering distinct progression trajectorie s and improving prognosis prediction.

Sp ecifically\, the T-Phenotype algorithm is adapted to the Lisbon ALS Clinic dataset\, leveraging phenotypic predictive clustering to group patients b y both outcome and disease trajectory similarities. Comprehensive preproce ssing strategies were developed to handle irregular clinical event timing\ , missing data\, variable-length time series\, and class imbalance. The al gorithm identified clinically meaningful phenotypes associated with the ne ed for NIV (endpoint C1) and PEG (endpoint C3)\, with optimal performance in shorter observation windows (6–12 months)\, revealing distinct risk p rofiles aligned with respiratory and bulbar function decline.

Co mbining endpoints into a multiclass task (C1+C3) highlighted challenges wi th class imbalance and reduced prediction performance\, although still pro duced interpretable phenotypes. Temporal analysis of cluster transitions i llustrated the model’s ability to capture diverse ALS progression patter ns dynamically\, emphasizing the disease'\;s heterogeneity and the lim itations of static classifications. Although the T-Phenotype model had sat isfactory results in predictive accuracy and clustering interpretability f or the binary endpoints (Hprc metric around 80%)\, its performance decline d for the combined endpoint (Hprc metric around 65%)\, its performance dec lined for the combined endpoint. Additionally\, it lacked sensitivity to c linical improvements post-intervention\, underscoring the need for richer datasets and methodological rigor.

The p otential of ML-based temporal phenotyping was demonstrated as a valuable t ool for understanding ALS progression and supporting personalized prognosi s and intervention planning. By integrating temporal dynamics into patient stratification\, this approach advances data-driven ALS monitoring and hi ghlights critical considerations for future research in this domain.

END:VEVENT END:VCALENDAR