What other descriptors can potentially use (beside bispectrum component) to train neural network to predict electronic structure of a material?
A non exhaustive list of descriptors (e.g. ACSF, SOAP, MBTR…) together with their implementation in python can be found at: DScribe — DScribe 2.0.x documentation
A descriptor that was recently introduced (and which is not listed in the web page above) is called Atomic Cluster Expansion (see Ralf Drautz Phys. Rev. B 99 , 014104).
Researchers are also increasingly using graph neural network where no descriptors is needed just atom types and positions see for instance [2204.05249] Learning Local Equivariant Representations for Large-Scale Atomistic Dynamics)