Title:	Modeling Part-whole Hierarchies with Object-Centric Learning
DNr:	Berzelius-2025-327
Project Type:	LiU Berzelius
Principal Investigator:	Karl Henrik Johansson <kallej@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2025-10-01 – 2026-04-01
Classification:	10201
Homepage:	https://www.kth.se/profile/amkm
Keywords:

Abstract

Deep neural networks have achieved outstanding success in many tasks ranging from computer vision (Krizhevsky et al., 2012), natural language processing (Vaswani et al., 2017), playing games (Silver et al., 2016) and more recently protein folding (Jumper et al., 2021). With more and more applications being driven by deep learning it becomes more and more important to understand their inner workings and the key components driving their success. It facilitates model interpretability, guides effective architecture design, and enables troubleshooting, leading to improved performance, robustness, and informed decision-making in the rapidly evolving field of artificial intelligence. In this project we will continue focusing on understanding the key components of Object-centric learning, a recent subfield of deep learning which has already found widespread attention (Locatello et al., 2020; Singh et al., 2021; Löwe et al., 2023; Jiang et al., 2023; Didolkar et al., 2024). Object-centric learning emphasizes the explicit representation and understanding of individual entities within a scene, promoting structured information processing and enhancing model interpretability. Through this investigation, we aim to provide a comprehensive understanding of modeling part-whole hierarchies using object-centric models. By leveraging the capabilities of object-centric representation learning, our project seeks to optimize, adapt, and efficiently apply these models across a myriad of domains. Our approach explicitly captures relationships between objects and their constituent parts, potentially enhancing interpretability, improving robustness to distributional shifts, and promoting better generalization to unseen compositions, thereby advancing the state-of-the-art in this field. This project is the continuation of the project in the previous round and we respectfully request an extension of our existing Berzelius allocation to continue this ongoing project which requires additional compute to complete planned experiments and analyses. Alongside the main work on part–whole hierarchies, we propose another project on protein–protein binder design that explicitly targets high-affinity binders. High-affinity protein–protein binders are crucial for real applications (therapeutics, diagnostics), but today’s pipelines rarely optimize directly for true affinity: typical “success” is defined as the rate of successful binding but generally neglects binding affinity. Furthermore, common in-silico scores such as iPTM, pLDDT, and iPAE, do not predict binding affinities well, while available wet-lab binding datasets are scarce (Danneskiold-Samsøe et al., 2024). Current existing methods such as 1) Iterative hallucination (Pacesa et al., 2024), 2) diffusion models and inverse folding (Hayes et al., 2025), and 3) co-generation all-atom models (Team et al., 2025; Chen et al., 2025) looks promising but either do not perform well in terms of designing high-affinity binders, or fully lack such kind of validation. In this project, we first propose to curate a dataset of wetlab-validated binders with relative binding affinity scores. Then, we will steer an existing co-generation all-atom base model that jointly designs binder sequence and structure using this dataset, via supervised fine-tuning and, where beneficial, reinforcement learning-style objectives, to explicitly bias binder designs toward higher binding affinity.