BGPT: Author Review: Teng Liu

Explore by Goal

Quick Explanation Copied

Teng Liu — scientific strength (data-grounded)

Across the provided set of works, Teng Liu appears to contribute to quantitative experimental design and mechanism-focused biology (e.g., antimicrobial peptide pharmacodynamics, retinal optoretinography elastography, regulatory genomics at single-nucleotide resolution, CRISPR-based regulatory grammar models, and mechanistic cell-biology pathways such as PARG dePARylation). Evidence quality looks strongest where multiple orthogonal assays, proper controls, and explicit phenotyping/measurement are present. Key uncertainties remain where validation is limited (small cohorts, single in-vivo model, or limited orthogonal ground truth), or where reproducibility/data availability isn’t fully explicit in the provided excerpts.

Representative cited works: defensin NZL antimicrobial peptide optimization ; retinal light-evoked intrinsic actuation elastography ; single-nucleotide CRE regulatory grammar prediction via eScreen .

Long Explanation

Author Review — Teng Liu (science strength, skeptical & evidence-based)

Date: June 04, 2026 • Focus: scientific merit from the provided raw-data excerpts (no extra assumptions).

What is known from the provided evidence

Mechanism-oriented experimental work appears repeatedly: antimicrobial peptide pharmacodynamics and safety/stability .
Quantitative in-vivo imaging + modeling is present in retinal biomechanics: local phase-referencing AO-OCT and hybrid analytical/finite-element modeling to infer outer-retina stiffness differences between healthy and RP cohorts .
Large-scale, data-intensive regulatory genomics is represented by eScreen: training on a compendium of CRISPR noncoding screens, predicting CRE activity at single-nucleotide resolution, and validating with CRISPRi/knockout and base editing perturbations .
Cross-context biology and causality logic is suggested by mechanism-first perturbation studies (examples from the provided set include PARG dePARylation biology and engineered epigenetic control), though in the excerpted materials the strength depends on whether orthogonal validation and data availability are explicit .

Visualizations from provided raw excerpt data (where numbers were given)

Note: the plots below use only the numeric values present in your provided data block for each paper (no external values).

1) NZL antimicrobial peptide — MIC and time-kill (extracellular vs intracellular proxy)

Data from the NZL defensin study include MIC values (µg/mL and µM) for four strains and qualitative time-kill outcomes at 1×/2×/4× MIC.

2) Hydrogel–elastomer interfaces — interfacial toughness (robustness across elastomers; benzophenone effect)

Interfacial toughness values were explicitly provided for multiple elastomer matches and treatment states.

Scientific critique: strength, weaknesses, and blind spots (based on provided excerpts)

Strength signals

Multi-stage validation appears in multiple provided works: e.g., the NZL study combines in vitro potency (MIC, time-kill), mechanistic assays (permeabilization/localization/morphology), and in vivo outcomes (survival, organ translocation, cytokines, histology) with explicit stability/safety testing .
Model–data alignment: eScreen’s structure explicitly aims to decode regulatory grammar using large CRISPR screen training and validates with perturbation modalities (CRISPRi/KO and base editing), which directly probes functional causality at the regulatory level .
Quantitative instrumentation and computation are used rather than narrative interpretation alone: retinal elastography uses local phase referencing and FE/analytical coupling to estimate mechanical parameters and to interpret layer-specific displacement changes in RP vs controls .

Weakness / uncertainty signals

Translation and generalizability are often limited by the scope of validation cohorts/models. For NZL, the excerpt notes reliance on a single mouse infection model and a single human cell line, with translation to other species/contexts not tested .
In vivo sample sizes can constrain statistical power and subgroup interpretation. The retinal optoretinography elastography excerpt states healthy n=10 and RP n=5, which limits disease heterogeneity coverage .
Reproducibility depends on data/code availability. In the provided excerpts, eScreen’s raw data and code are available (strong reproducibility signal), but some other listed works in your dataset have availability described as “on request” or not explicitly linked in the excerpt, which reduces immediate auditability .
Model-based inference can be brittle. For retinal biomechanics, inferred stiffness depends on FE-derived scaling/assumptions; the excerpt notes small RP cohort and reliance on anchor stiffness scaling from FE .

Potential blind spots (what could change the story)

Failure-to-replicate risk rises when studies rely on a single in-vivo model or when key numeric details aren’t fully externally accessible. The NZL excerpt includes acknowledged textual inconsistencies and limited long-term safety .
Phenotype scope can be narrow for mechanism claims. For example, PARG dePARylation logic shows strong in-vitro dependence and screens for modifiers, but the excerpt states no in vivo validation is reported .
Cross-context extrapolation: eScreen’s validation is strong in multiple perturbation modalities, but translational relevance to in vivo physiology and cell states beyond the training/validation set remains a known boundary unless explicitly tested .

Science-action buttons (bespoke follow-ups on BGPT)

Feedback:

Updated: June 04, 2026