DeepEphys

Using machine learning for biomarker discovery in human iPSC neuronal networks

Abstract

The advent of induced pluripotent stem cell (iPSC) technology has provided an attractive avenue to study neurological disorders, such as Parkinson’s disease (PD) or amyotrophic lateral sclerosis (ALS), in living human tissue. Neuronal cultures, derived from patient iPSCs, retain the unique genetic signatures of their donors and hence allow the study of specific disease aspects that cannot be inferred from genetic or single-cell level investigations alone (Ardhanareeswaran et al. 2017). Pharmaceutical companies have started to use iPSCs as disease models for pre-clinical drug screenings and biobanks with patient-derived samples are currently being set up. High-density microelectrode arrays (HD-MEAs) provide a fitting methodology to record from iPSC-derived neurons at both, high spatial resolution and high throughput ​ (Obien et al. 2014)​ . Moreover, HD-MEAs allow for tracking of neuronal-network activity across development and to dissect neuronal function using pharmacological or genetic challenges. Combining human iPSC technology with HD-MEA recordings provides a state-of-the-art phenotypic screening platform, which could accelerate ​ in vitro drug discovery and help personalize treatment strategies ​ (Fink and Levine 2018)​ . Although, the field has realized the potential of human iPSC-derived neuronal cultures for biomarker discovery, there are currently no widely-accepted analytical tools or standardized assays available to thoroughly assess the functionality of iPSC-derived neurons. To address this need, we will use machine learning algorithms to infer essential features, i.e., biomarkers indicative of the disease state, in neuronal activity and connectivity that allow identification of disease phenotypes and evaluation of pharmacological interventions.
The overall aim of the project is to provide a toolkit for the systematic study of cellular and network phenotypes of neuronal cultures derived from human iPSCs, and to develop biological markers for neurological disorders based on HD-MEA electrophysiological recordings. Both, the data used for this project and the developed analytical tools will be made publicly available, thus providing a resource to researchers in this and related fields.

Read the article about this project on our blog:

Identifying Biomarkers of Parkinson’s Disease with Pluripotent Stem Cells

Description

  • The project aims at twofold:

    • A proof of prinicple that electrophysioligical features allow two discriminate between healthy and Parkinson’s disease iPSC cultures.

    • Establishing experimental ground truth for extracellular connectivity inference between neurons.

  • The SDSC has a supporting role in the project of the first aim and is leading in the validation pipeline for the connectivity analysis.

  • The project will provide evidence, whether iPSC could be potentially used for drug development for neurodegenerative disorders, such as Parkinson’s disease. Furthermore, it will leverage the MEA technology for establishing a connectivity validation pipeline, for benchmarking neural connectivity inference methods.

Goals:

This project has three specific goals:

  1. To develop a standardized pipeline for the analysis of electrophysiological recordings from iPSC-derived neuronal cultures. We will curate existing datasets and provide a framework for the collection of new datasets.
  2. To develop new statistical tools for the comparison of iPSC HD-MEA electrophysiological recordings and quantification of cellular and network phenotypes.
  3. To integrate the biomarker discovery process in the SDSC plattform and to allow other researchers to access and use HD-MEA data, analysis code, and inferred biomarkers.

Conference communications and workshops

  • Kim, T., Hornauer, P., Donner, C., Hierlemann, A., Borgwardt, K., Schröter, M., & Roqueiro, D. S. (2020). Comparison of connectivity inference algorithms for classification of neuronal cultures using graph kernels. In ECML PKDD Workshop on Machine Learning for Pharma and Healthcare Applications, PharmML https://doi.org/10.3929/ethz-b-000466325