K. Trifonova, M. Falk, M. Rouches, S. Vaikuntanathan, M. Elowitz, A. Murugan
Can a single cell learn without neurons?
Typically, we think that cells develop regulatory programs through evolution and, as such, have set responses for any environmental stimuli they have seen in their evolutionary history. In this theoretical work, we show that cells in structured environments can learn new regulatory programs within their lifetime using molecular networks with a dense web of interactions and rate-sensitive autoregulation.
Trifonova, K.*, Falk, M.*, Rouches, M., Vaikuntanathan, S., Elowitz, M., Murugan, A. Trainable computation in molecular networks. bioRxiv. 2025.
L.H. Delgado, L.A.M. Sager-Smith, K. Trifonova, D.A. Mazziotti
Can machine learning models trained on data allow us to overcome the computational scaling problem of simulating many-electron systems?
We developed a novel machine learning algorithm that approximates exact molecular energies which are costly to compute as a weighted average of approximate upper and lower bound energies calculated with less expensive methods. Our neural network learns from molecule's reduced density matrices, which contain information about violations of N-representability conditions.
Delgado-Granados, L.H.,* Sager-Smith, L.A.M.*, Trifonova, K.*, Mazziotti, D.A. Machine Learning of Two-Electron Reduced Density Matrices for Many-Body Problems. The Journal of Physical Chemistry Letters. 16 (9): 2231-2237, 2025.
* These authors contributed equally.