Author Review Workspace

Author Review: Yunlong Cao — Visual, Evidence-based Critique

Visual first: bibliometric snapshot, key-study scoreboard, and strengths/risks with primary inline sources from Cao's major recent works.

Key primary sources (core evidence)

• DeepCoV preprint (2025)
• DXP antibodies preprint (2026)

Evidence synthesis — strengths and concerns (visual cue boxes)

Detailed critical appraisal (claims vs evidence)

1. Claims of predictive power (DeepCoV): The preprint reports high Pearson correlations and earlier hotspot detection; however, as an in-silico model its real-world utility depends on representative surveillance sampling, the quality and temporal coverage of DMS phenotypes, and prospective external validation—none of which are yet independently reproduced. The authors themselves note sampling and epistasis limitations DeepCoV results & limits
2. Antibody engineering and structural claims (DXP-006/106): Sub-nanomolar binding and a 3.55 Å cryo-EM structure (PDB 23QE; EMD-69166) are robust technical data points that support a mechanistic epitope-blocking model; preclinical efficacy in disease models is encouraging but limited by cohort sizes (n=5 per group), species differences, and sponsor involvement. The authors disclose company/patent ties and provide PDB/EMDB deposition references, increasing transparency but not replacing independent replication DXP antibody experimental evidence
3. Translational readiness: Both projects are translationally ambitious (surveillance-impacting forecasting and a therapeutic antibody program). For clinical translation: required missing steps include large-animal toxicology/PK, independent replication of in vivo efficacy, blinded histopathology, comprehensive ADME, and for DeepCoV—prospective real-time validation on streaming surveillance data and external blind tests. The DXP paper itself notes those gaps and sponsor bias risks DXP limitations disclosed

Practical recommendations (for peer reviewers / collaborators)

• Request raw model code, training/validation splits, and a prospective DeepCoV test on blind time-windowed data; evaluate sensitivity to sampling bias and alternate epistasis-aware encoders.
• Insist on larger, randomized, and blinded preclinical cohorts for DXP in multiple models, independent ADME/tox panels, and non-company laboratory replication before translational claims.
• Disclose and audit conflicts: patents and company founding raise verification bar; independent third-party validation is essential to offset sponsor bias.

What would change my assessment?

Stronger evidence that would raise confidence: prospective DeepCoV forecasting success on new emergent variants in real-time, large-scale independent replication of DXP-006/106 in non-affiliated labs, and GLP toxicology/PK data showing safety margins. Conversely, inability to reproduce model performance or independent failure of DXP antibodies in blinded studies would lower confidence substantially.

Notes: this review strictly uses the supplied publication metadata and the two primary preprints cited above; claims about bibliometrics reflect the supplied numbers and possible name-disambiguation issues; all translational statements are constrained by the preclinical/computational nature of the cited work.