#28 · Structural bio & protein design

ML models span viral vaccine design, riboswitch detection, and dengue inhibitor screening

The dominant thread is learned or computational modeling applied to viral countermeasures at distinct stages. The mpox study uses ESM-2, MHCflurry, and GNN/RNN models to design three multi-epitope subunit vaccine constructs from eight viral targets. The Coxsackievirus B3 study applies a sparsely gated Mixture of Experts model to classify synthetic riboswitch performance from RNA sequence. The dengue study pairs a viral pseudotime model and single-cell infection classifier on high-content imaging to screen 960 repurposed compounds. The remaining two preprints analyze human β-defensin 1 variants and STING conformational dynamics using structural modeling and molecular dynamics, with STING also using a latent-space autoencoder, rather than viral targets. What is new is the coupling of learned models with design or readout across stages, from computational vaccine construction to phenotypic screening. All evidence is computational or screening-stage as described.