Quantified self projects
Apart of my main focus on computational neuroscience, I became interested in the domain of quantified self especially in the context of optimising ones health, healthspan and lifespan. Bellow you will find few possible projects from this domain, some of which are suitable as a software project or thesis.
-
Web tool for collaborative cataloging of optimal biomarker values.
While standard medicine typically only recognizes within-normal and out-of normal range of various biomarkers, extensive body of evidence exists on which biomarkers values predict the lowest mortality. There is a case to be made that optimising individuals biomarkers towards these optimal values would improve healthspan. Currently the issue is that this knowledge is largely implicit, and scattered across huge body of scientific literature. The goal of this project is to develop a web application that would allow collaborative curation of ranges of biomarkers. These are the key required features: (a) main screen with tabular depiction of the list of already curated biomarkers, with each associated with various extracted information; (b) data structures and associated UI for capturing the normal and optimal biomarker levels that have to allow for different ranges for different age groups, sex & race; (c) multi-user accounts and tracking of different user contributions. The list of data that should be possible to capture for each biomarker: (a) name; (b) description; (c) optimal and normal ranges each with reference and ideally an image of a figure from which the range has been derived. Internal tool for easy scraping/fiting of data/curves from imported images (which are typically figures extracted from papers) would be a nice bonus. -
Mobile application for habit tracking.
Mobile application for tracking daily habbits. Key features: (a) UI optimised for efficient input of data - minimal clickikng - time requirements; (b) data remains private for the user - is not collected in specialised central cloud backend; (c) but data is continuously/daily automatically exported to a general purpose storage account of the user - e.g. Google Sheets or Airtable. -
Machine learning methods for automatic detection of relationships in high-dimensional time series.
The goal of this project is to develop method that can take a time series of measurements from a given subject (data from one subject N=1 provided) and can identify relationships in the data. I.e. find which varibles impact which. Kek chalanges of this project are: (a) relatively low amount of data; (b) ability to detect spurious correlations. (c) need to deal with missing data points. The student is free to propose his own solution, but I have an architecture revolving around very shallow (2-3 layers) ANN with a crucial set of regularization in mind if you need guidance. -
Causality detection in high-dimensional human health and lifestyle time series.
Judea Pearl introduced a influential theory on proper handling of causality in probabilistic world. In this project you would apply this theory on high-dimensional human health and lifestyle time series to detect the causality among the measured variables - i.e. detect which varables causaly impact others.` -
Mortality or biologicla-age predictors from public data sources.
OA If interested in this topic come to talk to me about the details