Precise integration, derivation and prediction of experimentally obtained functional maps via the computation of self-organization model.

Abstract

An ideal scientific theory should accurately compute the real-world data, and seamlessly integrate, derive and predict the experimental outcomes with a generalized simulation framework. However, in neuroscience - particularly in studying functional maps of the visual cortex - a significant gap persists due to individual diversity: each animal exhibits a unique map pattern, complicating the development of a universal computational model. In this study, we demonstrate that a Kohonen self-organizing model (SOM) bridges this gap through its intrinsic determinacy. Early-stage, coarse functional maps inherently determine their final mature patterns and can serve as scaffolds to fully derive complete maps. By integrating experimentally obtained functional maps as initial feature vectors (scaffolds), we achieved stable "transplanted" maps across iterations, despite high inter-individual diversity (tested in different monkeys, cats, and ferrets). SOM derivation was robust: only 2% of experimental data could derive the whole functional map; errors (low resolution / blank mask / mismatched / blood vessel artifacts / poor signal-to-noise ratio) in experimental maps could be corrected in the model. Importantly, the scaffold-defined maps retained plasticity, enabling precise computational predictions. For example, orientation adaptation effects - simulated from pre-adaptation map from a cat - were later confirmed by real-world adaptation experiments in the same cat. This tight experiment-simulation integration validates how basic self-organization principles govern real functional maps, reconciles previously conflicting experimental observations, and exemplifies a unified framework for neuroscience theory and experimentation.