Machine learning in microbiome post-doc (EXTENDED DEADLINE!)

Jagiellonian University
Małopolska Centre of Biotechnology
Poland Kraków


# Predicting microbiome changes over time and upon perturbation: applications of machine learning to microbiome research and therapeutic strategies

The human body forms different ecological niches for microoranisms. We host a plethora of microbes on each part of our body – different parts of our skin, our mouths; and in our body, especially in the gut. Those different niches are called microbiomes, for example the gut microbiome. In a bit over a decade the scientists described different microbiomes and learned that the microbiome is not only necessary, as it co-evolved with us, diverse and complex, but also important to health. Changes in the composition of microbes may result in an imbalance within the ecosystem of our guts called dysbiosis, which in turn may lead or contribute to disease. We already know a wide array of microbiome-disease associations, from the quite expected ones, such as obesity or gastrointestinal problems (ulcerative colitis or Crohn’s disease), to diseases presumed far removed from the gut – type-1 diabetes, allergies, depression, anxiety, neurodegenerative disorders, cancer, and others.

With those significant health implications and vast amounts of data gathered over the years we are now in a position to turn the knowledge into action – learn to predict how the microbiome changes over time and upon perturbation. With this, we are going to learn how to change the microbiome in order to promote health or devise therapeutic strategies. Those strategies may involve probiotic interventions, dietary or lifestyle changes, not necessarily classical small-molecule drugs. This is no easy task as the microbiome is massive – 10s of trillions of microbial cells representing 100s of microbial species, harboring over 100 times more genes than the human genome. On top of that the microbiome is dynamic. Unlike our genome, which is fixed from birth, the microbiome changes from day to day, evolves with our diet, lifestyle and other life events.

In this project we will employ a cautious step-wise strategy to pave the way for future smart microbiome-oriented therapies. As a first step, we will construct computational models of how the microbiome changes from day-to-day. Analyzing long time-series data from several individuals we are going to construct statistical and machine learning-based models (auto-encoders) to predict those changes. Thanks to those models, we will learn what features (microbes or combinations of microbes) are important for microbiome evolution over time. We will also better learn which specific algorithms and neural network architectures work for this problem. Once we master this task, we will be ready to proceed to a more complex one.

One of the few microbiome-oriented therapeutic strategies are the fecal microbiome transplants (FMTs). Initially, they were used for the treatment of Clostridium difficile infections, but now they also find their use in the treatment of ulcerative colitis. Collecting the data from already published studies, we will predict the composition of patient’s microbiome after an FMT. This is a substantially more complicated task, so we will use previously designed models for time-series data to give us a head start (transfer learning paradigm).

As a result of this project, we will learn how to predict the human gut microbiome changes over time and in response to FMTs. This will be a first step towards intelligent microbiome-oriented therapeutic strategies which give significant hope for many of the most debilitating diseases of today.

#The role

Your role is going to be to develop the machine learning algorithms to predict the microbiome composition over time and upon perturbation. You are going to work closely with a PhD student who is going to curate and analyse the metagenomic microbiome data.

As a part of your position you are also going to be expected to contribute to the computational expertise within the group and to be proactive in solving other bioinformatic research tasks.

The position will be funded by National Science Centre SONATA 15 project.

PLEASE NOTE: due to NCN regulations I am not allowed to hire postdocs who obtained their PhDs at the Jagiellonian University. For more information on the reason behind that or to share your opinion, please contact the agency via e-mail::


# The successful candidate will
* have a PhD degree in computer science, or a quantitative field such as biology, bioinformatics, physics or mathematics,
* have excellent written and oral communication skills in English,
* have expertise in machine learning (esp. deep learning methods and software, e.g. TensorFlow, Keras),
* have documented experience in test-driven software development,
* have a collaborative mindset,
* enjoy interdisciplinary work,
* understand the importance of communication and interactions with other group members, collaborators and Centre members.

# The ideal candidate will
* fulfill all of the expectations of the “successful candidate”,
* have a track record of presenting at conferences and published research,
* have experience in data science, metagenomics, or biology,
* have an active interest in microbiome research.

Start date

September 01, 2020

How to Apply

Please send an email with a subject line “Sonata postdoc” to including:
* a cover letter including a short description of your key achievements, as well as a short explanation for why you would like to join my lab and how this position fits into your interests and career aspirations,
* CV,
* sample code or a link to a repositiory,
* names, contact details and professional relationship status of 3 potential referees.


Tomasz Kosciolek