Using Deep Learning Models Pretrained by Self-Supervised Learning for Protein Localization

TL;DR

This study demonstrates that self-supervised learning models pretrained on large datasets like ImageNet-1k and HPA FOV can effectively transfer to protein localization tasks in microscopy, even with minimal fine-tuning.

Contribution

It evaluates the generalizability of SSL models pretrained on domain-specific datasets for microscopy, highlighting their strong zero-shot and fine-tuned performance.

Findings

HPA FOV-pretrained DINO ViT achieves highest zero-shot macro F1 of 0.822.

Fine-tuning improves performance to macro F1 of 0.860.

Single-cell embeddings from HPA pretraining outperform others in k-NN tasks.

Abstract

Background: Task-specific microscopy datasets are often small, making it difficult to train deep learning models that learn robust features. While self-supervised learning (SSL) has shown promise through pretraining on large, domain-specific datasets, generalizability across datasets with differing staining protocols and channel configurations remains underexplored. We investigated the generalizability of SSL models pretrained on ImageNet-1k and HPA FOV, evaluating their embeddings on OpenCell with and without fine-tuning, two channel-mismatch strategies, and varying fine-tuning data fractions. We additionally analyzed single-cell embeddings on a labeled OpenCell subset. Result: DINO-based ViT backbones pretrained on HPA FOV or ImageNet-1k transfer well to OpenCell even without fine-tuning. The HPA FOV-pretrained model achieved the highest zero-shot performance (macro $F_1$ 0.822 $\pm$ 0.007). Fine-tuning further improved performance to 0.860 $\pm$ 0.013. At the single-cell level, the HPA single-cell-pretrained model achieved the highest k-nearest neighbor performance across all neighborhood sizes (macro $F_1$ $\geq$ 0.796). Conclusion: SSL methods like DINO, pretrained on large domain-relevant datasets, enable effective use of deep learning features for fine-tuning on small, task-specific microscopy datasets.