Learning Informative Features for Predicting Patient Outcome Through Mining Tumor-Specific Causal Networks (Yifan) & "You can tell by the way I use my walk." Predicting the presence of cognitive load with gait measurements.(Pritika)
Yifan's Abstract: Somatic genome alterations (SGAs) are one of the major causes of cancer. Among all the genetic alterations observed in a cancer cell, only a small fraction, known as drivers, directly contribute to tumor growth. The set of drivers in each tumor preserves most of the tumor-specific oncogenic information. With this information, personalized therapeutic schemes can be proposed, which is now known as precision oncology. The major obstacle for promoting the development of precision oncology is the high variability of driver sets across tumors; however, this variability can be resolved as the genes that carry driver alterations often integrate into only a few cellular signaling pathways. In this paper, we propose a methodology to learn cancer signaling pathway representative genetic features, which are informative for predicting patient outcomes. The genetic features we learned are in the form of clusters of differential expressed genes (DEG), identified based on the causal relationships between SGAs and DEGs inferred by the Tumor-specific Driver Identification (TDI) algorithm. The methodology was applied on breast cancer (BRCA) and Glioblastoma (GBM) data for evaluation. Seven and fifteen DEG clusters were identified for BRCA and GBM, respectively. Each DEG cluster represents the genetic outcome of a signaling pathway affected by up to three major drivers. When using the DEG clusters and clinical features as covariates in survival analysis, five BRCA and six GBM patient groups of distinct survival patterns were identified. The methodology can also be applied to other cancer types with minor modifications. The findings we obtained with the methodology provides insights in to the disease mechanisms for specific groups of patients, and can potentially expand the therapeutic choices for individual patients and promote precision oncology.
Pritika's Abstract: Machine learning on processed acceleration signals has shown that a healthy person’s gait can be affected by cognitive load. Dual tasking research on gait and cognitive load have the potential to elucidate motor control in aging and clinical populations. This study classified the presence of cognitive load based on gait accelerometry measurements of the walking patterns of ten healthy individuals (18-35 years old). Data points were collected using six triaxial accelerometer sensors and treadmill pressure reports. Stride and window extraction methods were used to process these data points; descriptive statistics of these strides and windows were used as features. We performed binary classification of cognitive load using logistic regression, support vector machine, random forest, and learning vector quantization. Within and between subjects, a cognitive load was predicted with accuracy values ranged from 0.93 - 1 by all four models. Various feature selection methods demonstrated that only 2-20 variables could be used to achieve similar levels of accuracies. Furthermore, coupling sensors with machine learning algorithms to detect changes in gait patterns between walking tasks, most of which are too subtle to identify with the human eye, may have a remarkable impact on the ability to detect potential neuromotor illnesses and fall risks. In doing so, we can open a new window to human health and safety prevention.