Imbalance between excitatory and inhibitory signals in the brain (i.e. E:I imbalance) has long been theorized as one of the primary pathophysiological explanations behind autism. While the E:I theory is amongst the most influential in the field, very little is still known regarding how applicable the theory is to most human patients with an autism diagnosis. Filling this gap is of the utmost importance for honing in on heterogeneous mechanisms that may affect different types of patients, but also for enabling further advancements in treatments that target E:I mechanisms. Here we tackle this issue by using computational and animal modeling techniques to better understand how E:I mechanisms can be inferred from techniques such as EEG and fMRI. Our computational models of recurrently connected excitatory and inhibitory neuronal populations can simulate realistic local field potential (LFP) and fMRI BOLD time-series readouts, while holding constant the ground truth of synaptic E:I ratio as well as a number of other E:I relevant parameters. We have developed a decoding model that allows us to use time-series ˘spectral barcodesˇ of neural time-series data (LFP, BOLD) to accurately predict ground truth estimates of synaptic E:I ratio and other E:I relevant parameters. In this work, we will build on these efforts to computationally model and decode E:I mechanisms, and then validate such models using animal model data where chemogenetic manipulations have been made to alter excitation or inhibition. We will then utilize these E:I decoding models to apply to n=19 genetic mouse models of autism with electrophysiological and fMRI data, with the aim to cluster and identify emergent subtypes of genetic causes of autism that differ in inferred E:I mechanisms. One of these mouse models of an autism-associated CNV at 16p11.2 will also be investigated with neuroimaging data from human patients with deletions or duplications in this CNV region (data from Simons Searchlight Collection). Scaling up our efforts to understand E:I mechanisms with neuroimaging data in human patients, we will further utilize E:I decoding models on large publicly available EEG and fMRI datasets of idiopathic autistic patients. Our aim here will be to describe effect sizes for case-control differences, describe how E:I mechanisms may heterogeneously affect different regions or networks of the brain, and to isolate data-driven E:I subtypes and develop a robust stratification tool that could identify such subtypes in new datasets. Finally, we will collect a new dataset of n=100 autistic children (5-18 years old) with 2 sessions (test and retest) of EEG and fMRI data collected under resting state and naturalistic movie viewing conditions. This dataset will allow us to examine test-retest reliability of E:I imaging biomarkers.