#4 · Structural bio & protein design

Computational tools for omics differential expression and structural similarity search

The dominant thread is computational infrastructure for interpreting large, heterogeneous molecular datasets. A meta-analysis of 20 single-cell datasets exceeding five million cells yields a cross-organ fibrosis atlas and open resource of shared fibrotic programs. CellDEEP addresses differential expression in scRNA-seq through metacell-based pooling with flexible strategies and reports favorable benchmarking. On the structural side, AlphaFind v2 provides similarity search over AlphaFold DB and TED domains using embedding pre-filtering followed by US-align refinement, while a cyclic peptide study uses ESM-2 embeddings with cyclic permutation averaging to select sequence libraries covering physical properties, demonstrated in a beta2-microglobulin application. The plasma EV study is heterogeneous to this thread, addressing pre-analytical isolation method selection rather than computational inference. What is new is the explicit design of pooling, embedding-based pre-filtering and structural refinement as central analytical choices.