Most human diseases exhibit a complex genetic etiology impacted by many genes and proteins in a large network of interactions. The process of evaluating gene-disease association through in-vivo experiments is both time-consuming and expensive. Thus, network-based computational methods capable of modeling the complex interplay between molecular components can lead to more targeted evaluation. In this paper, we propose and validate geneDRAGNN: a general data processing and machine learning methodology for exploiting information about gene-gene interaction networks for predicting gene-disease association. We demonstrate that information about the gene-gene interaction network can significantly improve the performance of gene-disease association prediction models. We apply this methodology to predicting gene-disease association for lung adenocarcinoma, a histological subtype of lung cancer, and perform genomic analysis on genes flagged as potential associations. 


geneDRAGNN - gene Disease PRioritizAtion using Graph Neural Networks

We present a novel model and approach, combining Transformers and VAE-like model that builds upon the established Akita/Basenji CNN architecture to enable exciting new possibilities of predicting 3D chromatin folding from DNA sequences. We establish the motivation for using self-attention and a simple yet feasible Bayesian extension, derived from the usual posterior predictive distribution. Finally, we analyse the model, demonstrate the benefits and novelty of our approach, and discuss meaningful next steps. 

Attention to Bayes: Self-Attention and Bayesian Methods for Predicting 3D Chromatin Folding From DNA Sequences

 Diabetes is a major global health issue, with cases predicted to rise from 451 million today to 642 million by 2040. We use three machine learning methods (k-NN, regression and neural networks) to predict the glucose levels of mice based on genetic expression data across five tissues. Based on the best-performing neural network model, we derive modules that correspond to metabolic pathways by retraining networks on permuted feature clusters. The neural networks performed the best of the three model families, achieving a mean absolute percentage error of 26.0\% on the adipose tissue. From the neural network models, we produce a list of 9 modules that have high impact in their respective models.

Machine Learning for Glucose Prediction to Identify Diabetes-related Metabolic Pathways

Inourstudy,webuildaframeworkbasedon graph neural network to predict the mental status of a twitter user. Experimental result has shown that our model obtained a stable performance in the real life scenario.

A Multi-Modal Approach for Depression Detection

Using meteorological data, time series forecasts of disease outbreaks can better
capture the true epidemiological profile of tropical diseases such as malaria. In this
study, several methods of time series analysis are employed to study the disease
patterns of malaria in the Indian state of Odisha. Weather information, including
temperature and precipitation data, is incorporated alongside monthly case numbers
in SARIMA and LSTM models. The viability of transferring the model trained on
malaria in Odisha to dengue in Bangkok, Thailand, is also explored. The methods
outlined in this paper can serve as the basis for forecasting mosquito-borne disease
outbreaks in settings with a poor data-collection infrastructure.

Modeling Malaria Outbreaks Utilizing Weather Factors

As life expectancies increase,  dementia and neurodegenerative diseases affect increasing portions of the population. Alzheimer’s disease (AD) in particular has shown increasing variation in disease progression which suggests subtyping as an avenue for improving understanding and treatment of the condition. Recent research in neuroimaging has revealed distinct phenotypes of AD but a genetic explanation of these subtypes has yet to be uncovered. To that end we attempt to classify these subtypes from SNPs. Multiple tree-based models demonstrate that we found no correlation between tau trajectories and clinically SNPs in a small sample of ADNI patients.  Beyond the possible confounds of small sample size, our results are also consistent with the lack of previous studies that implicate SNPs in tau pathology. Based on our discoveries, we include multiple suggestions for future studies into tau-disposition of AD as well as suggestions for linking SNP data to other kinds of phenotypic subtypes of AD.

Investigating Genotypic Definitions of Tau Trajectory Subtypes in Alzheimer’s Disease using Tree-Based Models

Mechanical ventilation is a key form of life support for patients with pulmonary impairment. Nonetheless, the optimal treatment regime is often unknown, leading to sub-optimal care and increased risks of complications. This work aims to develop a decision support tool to personalize mechanical ventilation. We present DeepVent, an off-policy deep reinforcement learning model that determines the best ventilator settings throughout a patient's stay. We evaluate our model using Fitted Q Evaluation, and show that it is predicted to outperform physicians. Moreover, we address the challenge of policy value overestimation in out-of-distribution settings using Conservative Q-Learning and show that it leads to safe recommendations for patients. We also design an intermediate reward based on the Apache II score to further improve our model's performance.

Personalization of Mechanical Ventilation Treatment using Deep Conservative Reinforcement Learning

Despite the worldwide effort to gain a  deeper comprehension of the molecular mechanism that characterizes cancer, its blueprint is still elusive. In the last decade, results in the context of cancer immunology helped researchers to identify a set of biological objects, namely antigens, which now represent a promising direction. Mass spectrometry is the typical investigation involved in collecting molecules. This method and modern techniques allow a limited but satisfactory throughout on peptide knowledge. However, the diversity coming out from such experiments is still poorly characterized. In this paper, UnChAnTies(Unsupervised characterization of antigen communities) is introduced. This tool is a computational platform that enables unsupervised clusterings and identifies critical variables characterizing each community of antigens.  The goal of the pipeline is to identify relevant structures through unsupervised feature extraction with a model-free approach.  Additionally,  shed light, that takes part in the interplay between various kinds of cancer through their antigen are determined

Immunopeptidome Cluster Profiling with Critical Variable Selection

The outbreak of the infectious and fatal disease COVID-19 has revealed that pandemics assail public health in two waves: first, from the contagion itself and second, from plagues of suspicion and stigma. Now, we have in our hands and on our phones an outbreak of moral controversy. Modern dependency on social media has not only facilitated access to the locations of vaccine clinics and testing sites but also—and more frequently—to the convoluted explanations of how “COVID-19 was a FIFA conspiracy” [1]. The MIT Media Lab finds that false news “diffuses significantly farther, faster, deeper, and more broadly than truth, in all categories of information, and by an order of magnitude” [2]. The question is, how does the spread of misinformation interact with a physical epidemic disease? In this paper, we estimate the extent to which misinformation has influenced the course of the COVID-19 pandemic using natural language processing models and provide a strategy to combat social media posts that are likely to cause widespread harm.

“COVID-19 was a FIFA conspiracy #curropt”: An Investigation into the Viral Spread of COVID-19 Misinformation

Traditional methods and clinical practice of medical image classification have reached their limit regarding performance, making it difficult to improve through normal means. The emergence of deep neural networks such as the convolutional neural network (CNN) have proven to be an effective method on varying image classification tasks. To improve the performance of these methods, we propose to use image enhancement techniques to improve the quality and perception of medical images and to reduce inherent noise. We propose to use two spatial domain image enhancement techniques as a preprocessing step, median filter, a filtering technique, and contrast limited adaptive histogram equalization, a contrast enhancement technique, on Covid19 CT scans before processing them through pretrained transfer learning algorithms. Through this experiment, we will demonstrate the effect of image enhancement on classifying and predicting Covid19 images and diagnosis.

University of Toronto 2021 ProjectX Submissions