Guardado en:
Detalles Bibliográficos
Autores principales: Civale, Vincenzo Yuto, Semeraro, Roberto, Bagdanov, Andrew David, Magi, Alberto
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2604.14838
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915940439949312
author Civale, Vincenzo Yuto
Semeraro, Roberto
Bagdanov, Andrew David
Magi, Alberto
author_facet Civale, Vincenzo Yuto
Semeraro, Roberto
Bagdanov, Andrew David
Magi, Alberto
contents Current single-cell foundation model benchmarks universally extract final layer embeddings, assuming these represent optimal feature spaces. We systematically evaluate layer-wise representations from scFoundation (100M parameters) and Tahoe-X1 (1.3B parameters) across trajectory inference and perturbation response prediction. Our analysis reveals that optimal layers are task-dependent (trajectory peaks at 60% depth, 31% above final layers) and context-dependent (perturbation optima shift 0-96% across T cell activation states). Notably, first-layer embeddings outperform all deeper layers in quiescent cells, challenging assumptions about hierarchical feature abstraction. These findings demonstrate that "where" to extract features matters as much as "what" the model learns, necessitating systematic layer evaluation tailored to biological task and cellular context rather than defaulting to final-layer embeddings.
format Preprint
id arxiv_https___arxiv_org_abs_2604_14838
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Intermediate Layers Encode Optimal Biological Representations in Single-Cell Foundation Models
Civale, Vincenzo Yuto
Semeraro, Roberto
Bagdanov, Andrew David
Magi, Alberto
Artificial Intelligence
92B20, 68T07
J.3
Current single-cell foundation model benchmarks universally extract final layer embeddings, assuming these represent optimal feature spaces. We systematically evaluate layer-wise representations from scFoundation (100M parameters) and Tahoe-X1 (1.3B parameters) across trajectory inference and perturbation response prediction. Our analysis reveals that optimal layers are task-dependent (trajectory peaks at 60% depth, 31% above final layers) and context-dependent (perturbation optima shift 0-96% across T cell activation states). Notably, first-layer embeddings outperform all deeper layers in quiescent cells, challenging assumptions about hierarchical feature abstraction. These findings demonstrate that "where" to extract features matters as much as "what" the model learns, necessitating systematic layer evaluation tailored to biological task and cellular context rather than defaulting to final-layer embeddings.
title Intermediate Layers Encode Optimal Biological Representations in Single-Cell Foundation Models
topic Artificial Intelligence
92B20, 68T07
J.3
url https://arxiv.org/abs/2604.14838