Identification and physical characterisation of sarcomere pattern formation using supervised machine learning - Master thesis

Aug 3, 2026

Upload by Ezra

The thesis addresses the challenge of analysing large volumes of fluorescence microscope images produced during biologists’ studies of myofibrillogenesis in Drosophila pupae. Supervised machine learning methods—Support Vector Machines and Random Forest—are applied to identify sarcomeres, the highly regular structural units that compose myofibrils. For MATLAB implementation, feature engineering leverages physical properties of the experimental system. Class imbalance between positive and negative examples is mitigated by a redefined learning cost, enabling more reliable classification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages109

Reading Time4 h 35 min

AccessFree to download

Introduction

Supervised machine learning

The concept of supervised learning

Popular learning algorithms: Support Vector Machines and Random Forest

Performance metrics

Applying supervised machine learning to sarcomere identification

Overall code structure

Input data

Structuring the data

Normalisation

Feature engineering

Countering class imbalance

Parameter screening

Results

Algorithms: Random Forest performs better than SVM

Subsets by sarcomere length: performance is improved

Models based on different timepoints: interchangeable to some extent

Long sarcomere identification: neighbourhood information is useful

Features: intensity data and sarcomere orientation are most important

Discussion and Outlook

Appendix

References

Technische Universit¨at Dresden Physics of Life Bereich Mathematik und Naturwissenschaften Biological Algorithms Group Fakult¨at Physik Masterarbeit zur Erlangung des Hochschulgrades Master of Science im Studiengang Physik Identification and physical characterisation of sarcomere pattern formation using supervised machine learning vorgelegt von Leon Sbosny geboren am 8.9.1996 in Chemnitz eingereicht am 17.10.2022 Erstgutachter: Prof. Dr. Benjamin M. Friedrich Zweitgutachter: Prof. Dr. Marc Timme Abstract To analyse the large amounts of image data that are generated by biologists with modern microscopes, machine learning algorithms became increasingly popular. In collaboration with Frank Schnorrer and Cl´ement Rodier at Institut de Biologie du D´eveloppement de Marseille, as well as Ian Estabrook at Physics of Life, TU Dresden, this thesis applies the supervised machine learning algorithms ‘Support Vector Machine’and ‘Random Forest’ to data obtained from fluorescence microscope images of myofibrillogenesis in Drosophila pupae with the aim to identify sarcomeres, the structures that makeup the highly regular myofibrils. For the implementation in MATLAB, methods such as ‘feature engineering’ are used to increase the performance by reinterpreting the input data and using physical characteristics of the sample system. The project also identifies the problem of class imbalance between positive and negative examples in the input data and counters it with a redefined learning cost. In conclusion, the use of machine learning algorithms for image analysis in biophysics is a very promising way to reduce manual labour. The choice of the best learning algorithm depends on the purpose the obtained output data should serve. ¨Uberblick Zur Auswertung der großen Mengen an Bilddaten, die Biologen mit heutigen Mikroskopenerzeugen, sind Methoden des maschinellen Lernens sehr interessant geworden. In Kollaboration mit Frank Schnorrer und Cl´ement Rodier vom Institut de Biologie du D´eveloppement de Marseille sowie Ian Estabrook von Physics of Life, TU Dresden, werden im Rahmen dieser Arbeit die ¨uberwachten Lernalgorithmen “Support Vector Machine”und “Random Forest” auf Daten angewendet, die aus Fluoreszenzmikroskopaufnahmen zur Bildung von Muskelfibrillen in Puppen von Drosophila gewonnen wurden. Das Zielist dabei, Sarkomere, die regelm¨aßig angeordneten Baueinheiten der Muskelfibrillen, zu erkennen. Bei der Umsetzung in MATLAB wird zur Erh¨ohung der Effizienz des Lernalgorithmus die Methode des “Feature Engineering” verwendet, um die Eingangsdaten unter Ber¨ucksichtigung der physikalischen Beschaffenheit des experimentellen Systems neu zu interpretieren. Zus¨atzlich besch¨aftigt sich das Projekt mit dem Problem ungleicher Anzahl von positiven und negativen Lernbeispielen, was durch Neudefinition der Lernkosten ausgeglichen wird. Insgesamt verspricht maschinelles Lernen erhebliche Arbeitserleichterungen bei der Bildauswertung in der Biophysik. Dabei kommt es bei der Wahl geeigneter Lernalgorithmendarauf an, wof¨ur die Ausgabedaten verwendet werden sollen. Selbstst¨andigkeitserkl¨arung Hiermit erkl¨are ich, dass ich diese Arbeit im Rahmen der Betreuung am Exzellenzcluster Physics of Life, TU Dresden, ohne unzul¨assige Hilfe Dritter verfasst und alle Quellen als solche gekennzeichnet habe. Ort, Datum Leon Sbosny Acknowledgements At this point I would like to thank Benjamin Friedrich and Ian Estabrook at Physics of Life, TU Dresden, for the opportunity to work on this project. Without the regular encouragement from Ben to try out new approaches and to look atthe subject from different perspectives as well as the constant support from Ian who introduced me to his research on myofibrillogenesis, this thesis would not have been possible. I also thank Frank Schnorrer and Cl´ement Rodier at Institut de Biologie du D´eveloppement de Marseille for providing the experimental data this project is ba

FAQs

Which supervised learning algorithms are used to identify sarcomeres?

The thesis applies Support Vector Machines and Random Forest to fluorescence microscope image data to detect sarcomeres.

How is MATLAB performance improved for the learning models?

Feature engineering is used to reinterpret the input data by incorporating physical characteristics of the experimental system.

How does the project handle class imbalance in training data?

Class imbalance between positive and negative examples is countered by redefining the learning cost used during training.

Aug 3, 2026

Upload by Ezra