From Raman to infrared: extending DIVA to FTIR spectra for quantitative seed phenotyping
Anoop C. Patil, Sreelatha Sarangapani, Yu-Wei Chang, Kasi Viswanath Kotapati, Ganga Sravanthi Cheerlavancha, Joana B. Pereira, Raju Cheerlavancha, Mervin Chun-Yi Ang, Gajendra P. Singh, Rajani Sarojam, Giovanni Volpe
Date: 24 August 2026
Time: 1:45 PM – 2:00 PM PDT
Place: Conv. Ctr. Room 2
Vibrational spectroscopy enables non-destructive probing of biomolecular composition in biological systems. We extend our Deep-learning Investigation of Vibrational spectra (DIVA) framework, originally developed for Raman spectral analysis, to Fourier-transform infrared (FTIR) spectroscopy for seed characterization. The pipeline integrates absorbance normalization, first-derivative transformation, and variational autoencoder (VAE)–based latent representation learning to capture structured spectral variability across wildtype and treated seed categories. Consensus derivative bands and pairwise spectral differences are extracted to identify discriminative biomolecular features between treated and wild-type seeds. Latent-space embeddings enable visualization of cluster separability and support classification. This work demonstrates that DIVA provides a scalable framework for quantitative FTIR analysis, enabling robust unbiased assessment of seed condition.