Hephaistos is an early stage startup that aims to revolutionize enzyme engineering field. Our machine learning algorithm creates novel biocatalysts to optimize resource use in the pharmaceutical and chemical industries. Our enzymes aim to substitute complex chemical processes that are inefficient and enable new reactions that are considered impossible to conduct.Forming part of a young team of six, I personally lead internal meetings and processes, conduct market analysis, reach out to potential customers and investors, discover and apply for funding opportunities, support our developer's team when needed.

Continuing the development of the Large Language Model tool for toxicological data extraction created during the internship, as part of translational Modeling and Simulation (tM&S) capability. The main focus lies applying previously developed extraction pipeline to different case studies within Roche and explore its potential improvements and limitations based on these real world scenarios. In addition, alternative LLM and data extraction approaches are evaluate to increase the versatility of the tool.Supervisor: Tatyana Y. Doktorova.

Establishing an extraction pipeline for toxicity related information from historical records and other internal documents using Large Language Models (LLMs), in particular the retriever augmented generation (RAG). Computational toxicology project as a part of the Predictive Modelling and Data Analytics (PMDA) chapter at pRED.Supervisors: Tatyana Y. Doktorova, Julia Pletz, Alessandro Brigo.

The thesis focuses on understanding the principles of the epithelial organization in 3D, which is relevant for tissue growth in embryonic development. The goal is to improve and extend SimuCell3D software (C++) to enable simulation of stable epithelial vesicle fusion and tube growth. Finally, the simulation results are compared to the theoretical data and the 3D imaging samples of the real tissue.In addition, the extension of the tool requires in depth literature research as well as participation in the peer review process of most recent publications in this area.Supervisors: Steve Runser, Dr. Roman Vetter, Prof. Dagmar Iber

The aim of the project is to set up a computational pipeline to improve the developability of antibodies. An autoencoder architecture, based on known protein language models (PLMs), was trained to create a meaningful antibody representation (the latent space). This representation can then be combined with a retrieval augmented generation (RAG) step to create modified sequences with increased developability parameters. Supervisor: Lena ErlachLaboratory for Systems and Synthetic Immunology, Prof. Sai Reddy.

Modeling of an oral microbiome using agent-based framework MESA in Python.Supervisors: Christian Kronseder and Tino Töpper.

Creation of a genetic regulatory network (GRN) to describe exhaustion in CD8+ T-cells in the context of hepatitis C virus infection using single-cell RNA sequencing data. The vast majority of the project was programmed in R and some specific parts were coded with Python. Further applications and tools: Cytoscape, SCENIC, Random Forest, Bonesis.Supervisors: Prof. Carl Herrmann and Dr. Carlos Ramirez.

Creation of a database to store and analyze lab data. The main goal of the database was to facilitate and accelerate the information flow between the laboratory and the rest of the company in the most convenient and sustainable way. Database operates on SQL programming language.Supervisors: Thomas Mannheim and Leonie Boller.