Computational Science Degree Projects at COSHE
Can we interest you in a thesis project in computational science?
On this page we collect suggestions for projects covering artificial neural networks, systems biology, bionanophysics, quantum computing or complex statistical modelling.
Please scroll down for the newest projects
Students from physics, computational science and other disciplines are invited to do their thesis work with us at Computational Science for Health and Environment (COSHE). The COSHE theme was formed at CEC in 2023 by the members of Computational Biology and Biological Physics (CBBP, formerly at Astronomy and Theoretical Physics).
Both bachelor and master students are welcome to do their projects in one of our research groups. We offer a strongly interdisciplinary research environment where staff and students have a background in diverse subjects that include physics, computer science, medicine and mathematics and environmental science. Our thesis students are integrated into the group both scientifically and socially for the duration of the project, and gain valuable experiences and connections. Many of our master project students are subsequently recruited to PhD positions, either with us or elsewhere.
Below you will find descriptions of possible topics for your degree project. You can also contact the teachers directly and ask about possible projects. Many examples of earlier theses are available from LUP Student papers.
This does not apply for students in computational science and in envrionmental science. We currently do not have our own courses for bachelor or master projects, but act as external supervisors in collaboration with other departments. Thus, you contact us to discuss possible projects and the course coordinator for your regular degree project course (e.g. FYSK03, FYSM33 or BERM01). The course coordinator needs to determine that the project fits within your main subject and, if needed, will also help you find a formal supervisor in their department.
Project descriptions
Quantum computing: Folding and design of toy proteins
Simplified lattice-based models provide a means for exploring the basics of the sequence–structure relationship of proteins. One then needs to solve discrete optimization problems, which, despite the simplicity of the models, become computationally challenging for large proteins. Quantum computing offers a potentially fast approach to difficult optimization problems. The aim of this project is to explore quantum computing based methods for solving lattice protein problems.
Contact/supervisor: Anders Irbäck (anders.irback@cec.lu.se)
Recommended knowledge: Statistical mechanics corresponding to FYTN15 and programming experience. Knowledge in theoretical biophysics and computational physics (corresponding to FYTN05/BERN05 and FYTN03, respectively) is helpful but not required.
Language: English or Swedish.
Bionanophysics: Parameter inference from single-molecule trajectories
Single-molecule experiments, such as single particle tracking
(SPT), enables precise quantitative analysis beyond ensemble averages. However, when analyzing experimental SPT trajectories, extracting kinetic rates and diffusion constants is made difficult due to different types of experimental noise. Two common such noise types are: motion blur (the molecules move during each sampling time) and localization noise (not enough photons were detected for accurate precision).
In this project you will dwelve into using Bayesian data anaysis for biomolecule trajectories. You will (i) simulate realistic movies of moving biomolecules as obtained in wide-field fluorescence microscopy setups, (ii) from the movies extract trajectories, and (iii) from the trajectories use state-of-the-art Bayesian methods for estimating parameters. You will include all experimental noise components, and address how including or neglecting these affects your parameter estimates.
Contact/supervisor: Tobias Ambjörnsson (tobias.ambjornsson@cec.lu.se)
Recommended knowledge: Computational physics corresponding to FYTN03 (Computational Physics) and programming experience.
Language: English or Swedish.
Bionanophysics: DNA melting in forensics
The sequence of base pairs of DNA molecules is the genetic blueprint of any living being. A DNA molecule forms a double-helix where the complementary interactions between bases strike a balance between stability (protecting the integrity of the double helix) and accessibility (for instance, the bases need to be accessible when reading the DNA sequence). Because of the non-covalent nature of the bonds between bases, the DNA molecules “melt” (base pairs get separated) when heated above room temperature. The associated melting curve (fraction of “melted” DNA base pairs as a function of temperature) is sensitive to DNA sequence, with local melting temperatures that range from 60 to 110 °C. The sensitivity of DNA melting to DNA sequence can be used in DNA fingerprinting. By performing optical measurements, the fraction of melting base pairs as a function of temperature can be extracted. Since DNA melting curves are sensitive to DNA sequence, DNA fingerprinting based on DNA melting has the potential to complement standard DNA sequencing.
As an example, DNA molecules are subject to a process called methylation, where base pairs undergo chemical modifications. These modifications alter the interaction energy between bases and, as a consequence, the DNA melting curves are sensitive to the methylation fraction and position of the methylation events. Since methylation events are dependent on lifestyle and age as well as random contributions, two individuals with identical DNA sequences (twins) will have different methylation patterns. Therefore, the associated DNA melting curves will differ, a fact which has been used in forensics.
In this project you will numerically investigate the effect of methylation with an established model for DNA melting (the Poland–Scheraga model). The work will be performed in collaboration with theoretical and experimental partners at Umeå and Uppsala Universities.
Contact/supervisor: Tobias Ambjörnsson (tobias.ambjornsson@cec.lu.se)
Recommended knowledge: Good programming experience and knowledge of statistical mechanics corresponding to FYTN05/BERN05.
Language: English or Swedish.
Machine learning: Microplastics identification
Microplastics are small particles of plastics, ranging in size from micrometers to millimeters. As the wide range of plastics used industrially and in households are shed and broken down, large quantities of microplastics end up in our waste. European farmland is estimated to receive 31,000–42,000 tons annually from biosolids and composts, and this is a growing concern for both human health and then environment.
Many machine learning methods have been applied to spectroscopic and imaging data for the purpose of identification and quantification of microplastics. We have a dataset where biological waste has been treated to break down organic materials, and the remaining solids have been filtered and imaged. In these microscopy images, particles that may be plastics can be seen and counted, and IR spectroscopy methods can be used to determine what types of plastics they are.
It is desirable to automate the task of identifying potential plastics particles, but this is not trivial because of the complex structure of the filters themselves. The image to the right here shows an example of unsuccessful image segmentation by a convolutional neural network; the areas marked by the network contain nothing. Your task in this project is to develop and train a neural network model that successfully finds regions/points that are suitable candidates for further exploration by spectroscopic measurements.
Contact/supervisor: Carl Troein (carl.troein@cec.lu.se)
Recommended knowledge: Programming skills and knowledge of artificial neural networks and deep learning corresponding to FYTN14/BERN04.
Language: English or Swedish.
Machine learning: Explainable neural networks
An artificial neural network (ANN) is a powerful and flexible machine learning tool. It has gained popularity in recent years due to its connection to deep learning and its success in e.g. image analysis and language processing. When using neural networks for e.g. medical diagnosis or outcome prediction, it is of vital importance to be able to explain the reasoning behind an advice. Generally this is difficult for non-linear machine learning methods. This topic of unmasking the “black boxes” has become even more important with the present focus on deep learning.
In this project you will explore different methods for explaining the operation of complex neural networks models.
Contact/supervisor: Mattias Ohlsson (mattias.ohlsson@cec.lu.se)
Recommended knowledge: Programming skills. Also it is advisable to have knowledge of artificial neural networks and deep learning corresponding to FYTN14/BERN04.
Language: English or Swedish.
Machine learning: Methods for Predicting Dementia and Parkinson's Disease Evolution
Alpha-synuclein aggregation in neurons or glia is considered the central pathology of Parkinson’s disease (PD), Lewy Body Dementia (LBD), Multisytem atrophy (MSA).
Alpha-synuclein measurements in skin biopsies represent a reliable way to get information about the disease. Hyperspectral imaging of skin biopsies is a very promising method to produce relevant imaging data that can be used for machine learning models. We aim to develop such models for investigating whether skin biopsies can be used to predict whether PD or LBD started first in the brain or in the gut. This is one project example where we focus on the start of the disease.
Moreover the PD, LBD and MSA have a start in REM-sleep behaviour disorder and body parts uncontrolled movement. Based on short videos of patients we aim to use machine learning methods for predicting stages of the disease.
The project is a collaboration with research groups at medical faculty that have a clinical background with acces to patients data or imaging backgrounds with access to facilities at MAX IV.
Contact/supervisor: Victor Olariu (victor.olariu@cec.lu.se) and Patrik Edén (patrik.eden@cec.lu.se)
Recommended knowledge: Neural networks and deep learning corresponding to FYTN14/BERN04, programming skills.
Language: English or Swedish.
Systems biology: Gene regulatory network models for cell reprogramming and differentiation
We offer a range of projects on developing computational models for gene networks that govern skin cell reprogramming to stem cells or direct reprogramming to specific types of neurons. For example, finding new better ways to produce transplantable neurons is highly relevant for patients that suffer from Parkinson’s or Alzheimer's diseases. Single cell data are available from our established collaborations with various wet labs.
These projects involve data analysis and model optimisation along with deterministic and stochastic model simulations to predict solutions for improving the production of target cells.
Contact/supervisor: Victor Olariu (victor.olariu@cec.lu.se)
Recommended knowledge: Programming skills and systems biology (corresponding to FYTN12/BERN06). It is also advisable to have knowledge in computational physics or computational science (corresponding to FYTN03 or BERN01).
Language: English or Swedish.
Systems biology: Production of plastic building blocks from sugar
The excessive use of single-use plastics has resulted in severe environmental, social and health consequences, partly because of their longevity in nature due to a lack of natural degradation systems. A promising solution is to develop a new generation of plastics that are bio-based and recyclable or compostable. Examples of raw materials with the potential to become building blocks for this new generation of plastics are 4-hydroxybutyrate (4-HB) and 1,4-butanediol (1,4-BDO). The production of environmentally friendly plastics from these biomaterials is hampered by their limited industrial production.
In this project, you will work with modelling microbial cell factories and improve our knowledge on how to increase bio-based production of the molecular building blocks 4-HB and/or 1,4-BDO. If you are interested in translating biological challenges into computational solutions and vice versa and have some knowledge of biochemistry, bioinformatics or computational sciences, this may be the project for you. The project is carried out in collaboration with the Division of Biotechnology at the Department of Chemistry.
Contact/supervisor: Victor Olariu (victor.olariu@cec.lu.se) and Carl Troein (carl.troein@cec.lu.se)
Recommended knowledge: Programming skills and systems biology (corresponding to FYTN12/BERN06).
Language: English.
Magnetic field and core flow variations over the past 9000 years
By studying geomagnetic field variations at the Earth's surface, we can learn more about the deep Earth where the field is generated. We offer a variety of projects on developing models of how the geomagnetic field and core surface flow have varied over the past 9000 years. These range from purely data-driven statistical geomagnetic field models based on different types of datasets, each associated with their unique challenges, to more physics driven core field and core flow models informed by statistics derived from numerical geodynamo simulations. Potential project aims include solving problems related to chronological data uncertainties through implementation of more efficient MCMC applications and exploring alternative model parameterizations for hypothesis testing.
Contact/supervisor: Andreas Nilsson (andreas.nilsson@geol.lu.se) and Anders Irbäck (anders.irback@cec.lu.se).
Language: English or Swedish.
Geophysics: Reconstructing solar activity and solar storms from ice core and tree ring data
Cosmogenic radionuclide records from ice cores (10Be, 36Cl) and tree rings (14C) provide information about solar and geomagnetic shielding of galactic cosmic rays far back into the past. However, such indirect “proxy data” is also influenced by the geochemical behaviour of these radionuclides such as e.g. climate and weather effects on their atmospheric transport. The challenge is to reliably extract the solar activity and solar storm information from noisy data. However, we can make use of the fact that we usually have parallel records from different ice cores and tree rings allowing us to increase the signal to noise. Furthermore, we can also use our knowledge about the typical behaviour of solar activity and known confounding effects on the radionuclides. Potential projects aim to combine such information into a statistical model to improve solar activity reconstructions and to push the limits of the identification of large solar storms in the past.
Contact/supervisor: Raimund Muscheler (raimund.muscheler@geol.lu.se), Anders Irbäck (anders.irback@cec.lu.se), and Ullrika Sahlin (ullrika.sahlin@cec.lu.se)
Language: English or Swedish.
AI-driven Holographic microscope to study marine plankton predator-prey interactions
The Signals in the Sea group is a cross-disciplinary research team (physics + biology) at the Biology Department at Lund University, working in close collaboration with the Soft Matter Lab in Gothenburg. We develop and apply computational and optical methods to tackle biological questions.
We developed an AI-driven holographic microscope that derives the position and biomass from holograms created by small organisms. This method is developed to study predator-prey interactions in marine plankton. Now, we are looking for a motivated master student who would like to build and implement the microscope setup at the Biology building in Lund and contribute to further method development.
We are looking for a student that is interested in combining experimental optics work with computational methods, and wants to work at the interface of physics, biology, and AI. The project can start as soon as possible, and can be tailored to 30, 45 or 60 ECTS depending on the student. Feel free to reach out if you want to find out more about the project!
Publication: Harshith Bachimanchi, Benjamin Midtvedt, Daniel Midtvedt, Erik Selander, Giovanni Volpe (2022) Microplankton life histories revealed by holographic microscopy and deep learning eLife 11:e79760, https://doi.org/10.7554/eLife.79760
Homepage: pelagial.se
Supervisors:
Ass. Prof. Erik Selander, Dept. Biology, Lund University Erik.Selander@biol.lu.se
Dr. Harshith Bachimanchi, Soft matter lab, Gothenburg University
PhD Candidate Isa Hendriks, Dept. Biology, Lund University Isabel.hendriks@biol.lu.se
Environmental management, optimal control and statistical calibration
Contact Ullrika if you are interested in a project where you work with solving dynamic control problems with applications on environmental management and policy. Focus can be on dynamic programming and/or on statistical calibration of the dynamic models. It can be theoretical models where you address numerical problems with the optimisation or problems where you use real data to calibrate the mathematical models. I have applications on management of human-wildlife conflicts, invasive alien species, biodiversity conservation, but are open for other suggestions. Ullrika Sahlin ullrika.sahlin@mgeo.lu.se
Biodiversity data from remote sensing
PhD student Carlos Jair Munoz has an idea about using neural networks to process flight images to create more refined data on species monitoring. carlos_jair.munoz@mgeo.lu.se
Challenges in uncertainty quantification in applied assessment problems
Scientific assessments are knowledge production processes calibrating assessment models by integrating diverse evidence, applying rigorous expert judgement to evidence and its uncertainties, and exploring futures and their connections to risk management or policy decisions. The methodological practices. Simulation studies demonstrating the performance of a proposed methodoogy compared to current practice, followed by application on case study. Current practice to assess if a critical thresholds of hazardous chemicals are exceeded based on small samples from chemical analysis challenged by a fully Bayesian analysis considering uncertainty and dependences in multi-chemical data. Current recommendations to derive highest reliable percentiles from occurence data of chemicals in food challenged by Bayesian statistical modelling accounting for censored data, sample size, and uncertainty in model structure (model averaging) using probability models suitable for percentile estimation. Ullrika Sahlin ullrika.sahlin@mgeo.lu.se