Dr Romy Lorenz | Towards a neurobiologically-derived cognitive taxonomy

The classic taxonomy of cognitive processes was developed largely blind to the functional organization of the brain; therefore, classic cognitive tasks tend to tap multiple cognitive processes that involve multiple brain networks. Resolving this many-to-many mapping problem between cognitive tasks and brain networks is practically intractable with standard fMRI methodology as only a small subset of all possible cognitive tasks can be tested. This is problematic, as studying only a fraction from the large task space can result in over-specified inferences about functional-anatomical mappings with a misleadingly narrow function being proposed as the definitive role of a network, concealing the broader role each network may play in cognition.

In this talk, I present an alternative approach that resolves these problems by combining real-time fMRI with a branch of machine learning, Bayesian optimization. Neuroadaptive Bayesian optimization is a powerful strategy to efficiently explore more experimental conditions than is currently possible with standard methodology. I will present results from a study where we used this method to identify the exact cognitive task conditions that optimally dissociate frontoparietal brain networks. Our findings deviate from previous meta-analyses and hypothesized functional labels for these frontoparietal brain networks. Taken together the results form the starting point for a neurobiologically-derived cognitive taxonomy.

In addition, I will touch on the potential of the approach in combination with non-invasive brain stimulation (e.g., tACS) and for accelerated biomarker discovery. Interestingly, Bayesian optimization can also be combined with preregistration to cover exploration, mitigating researcher bias more broadly and improving reproducibility.

As I have just joined the MPI two months ago, I will also share my research vision going forward and will illustrate how automated meta-analytic and text mining techniques can help me in my endeavour to map associations between cognitive processes and brain networks as well as how deep learning methods could help us to move beyond descriptive mappings and advance mechanistic discovery in network neuroscience.

Poster