Review Papers Claim by Claim

Concise appraisal

The preprint reports a scalable pipeline that couples de novo binder design (RF Diffusion ProteinMPNN AlphaFold2) with LABEL-seq to functionally screen ~10k designed effector handles in human cells and identifies hundreds of reproducible degraders and stabilizers, with orthogonal validation including endogenous MCL1 degradation and mitochondrial reorganization (strong internal concordance and crosscell validation) Multiplex proteome editing preprint

Detailed review and critique

Paper summary (what they did)

The authors built a 10k scale library of de novo designed 100 aa effector binders (10,225 designs, 9,715 retained after barcode/representation filters) and tested whether each design, when presented as an effector handle fused to a constant ALFA target binder, altered intracellular abundance of a FLAG tagged GCK reporter using LABEL-seq (FLAG versus Myc IP of circular RNA barcodes) in engineered HEK293 landing pad cells; effects were analyzed across two biological replicates and validated with orthogonal reporter assays, Western blots for endogenous MCL1, apoptosis assays, and imaging for mitochondrial perturbation LABEL-seq multiplex screening description

Key results (data driven)

• Library and retention: Designed 10,225 candidate effector handles (100 aa) and after barcode/representation filtering retained 9,715 designs for analysis (95%) Library retention statistics
• Screen metric and reproducibility: Primary metric was the ratio of FLAG to Myc circRNA barcode read proportions; replicate correlation of per-design median was Pearson r = 0.72, Spearman rho = 0.67 across two biological replicates Replicate correlation
• Hit counts: In combined data 5,299/9,715 (55%) of designs significantly stabilized or destabilized the target (Wilcoxon paired test, BH corrected); specifically authors highlight 277 degraders and 204 stabilizers as reproducible functional handles across effectors Hit counts and statistics
• Orthogonal validation: Singleton GFP:GCK:ALFA reporter flow cytometry validations in HEK293T and HepG2 showed high concordance with LABEL-seq effects (Pearson r ≈ 0.93; Spearman ρ ≈ 0.98 in HEK293T; cross-cell r ≈ 0.91) and 17/18 selected designs validated directionally in orthogonal assays Orthogonal validation concordance
• Endogenous target and phenotype: Autophagy-related effector handles (GABARAP, LC3A) reduced endogenous MCL1 levels (to ~50% and ~27% of control) in MDA-MB-231 cells, and induced apoptosis/cell death consistent with functional consequences of degradation; autophagy effectors also perturbed mitochondrial networks when redirected to the outer mitochondrial membrane Endogenous MCL1 and mitochondrial phenotypes

Strengths

1. Scale and integration: The paper couples modern de novo design pipelines with a true cell-based, sequencing readout (LABEL-seq), enabling functional screening at orders of magnitude beyond traditional single-target biochemical assays Framework summary
2. Robust validation: Multiple orthogonal validations (flow cytometry reporters, endogenous Western blots, apoptosis assays, imaging) strengthen the functional claims beyond sequencing enrichments Validation description
3. Openable toolbox: Demonstrates both UPS and autophagy handles can be designed and screened, and shows functional outcomes (stabilization degradation organelle remodeling), expanding actionable effector space beyond canonical PROTAC ligases UPS and autophagy handles

Limitations and potential blindspots

• Single primary target for screen: Primary LABEL-seq screen used an ALFA tagged GCK reporter (GCK has a long half-life) — authors acknowledge GCK half-life likely limited dynamic range, so many true effectors might be under-called or effect sizes attenuated GCK half-life limitation
• Context dependence and portability: Several designs active against GCK reporter did not translate to endogenous MCL1 for UPS effectors — shows effector efficacy depends on target identity, cellular context, intrinsic degradation pathways; generalizability to all proteins/cell types remains unproven Context dependence observation
• Dropout/toxicity: ~2% of designs dropped out upon recombination; binders to some effectors (DUBs) had higher dropout rates linked to toxicity — this biases which designs survive and may enrich for non-toxic sequence families while discarding potent but toxic candidates Design dropout and toxicity
• Effect size distribution: Most hits had modest magnitudes; only a minority produced >50% degradation; mechanistic determinants (affinity kinetics intracellular stability localization) are not yet resolved — limits predictive transfer from design metrics alone Effect size heterogeneity

Methodological and statistical rigor

The authors used conservative non-parametric paired tests (Wilcoxon signed-rank) comparing FLAG vs Myc barcode proportions across barcodes per binder, retained only designs with >=3 barcodes and barcode frequency thresholds, and used BH correction for multiple testing — these choices reduce false positives but may reduce sensitivity to weak but real effects. Authors also performed permutation checks showing excess small p-values in the real data, strengthening evidence that results are not artifacts of pipeline Statistical controls and permutation

Reproducibility and data sharing

Methods are detailed (library cloning, subassembly, recombination, IPs, RT-PCR, sequencing, barcode mapping, statistics), and authors state raw and processed sequencing data are being uploaded to GEO with tables included in supplement, supporting reproducibility; code for barcode mapping is described (Python) though public repository link not listed in the excerpts — releasing code and raw FASTQ/lookup tables will be essential for independent reproduction and meta-analyses Barcode computational pipeline

Practical implications and future directions

• Utility: Provides a scalable route to generate diverse effector handles beyond small-molecule ligands and could expand the toolkit for programmable degradation/stabilization and subcellular remodeling (organelle sequestration) Framework and applicability
• Next steps: Systematic screens across multiple reporter proteins (short/long half-lives, different compartments), expanded effector panels, deep learning models trained on these functional datasets to predict intracellular activity, and mechanistic studies (measuring binding kinetics intracellular half-life, coIP of native complexes) are logical and necessary follow-ups Future mechanistic and modeling work

Bottom line assessment

The study is high-quality, methodologically careful, and meaningfully advances the field by demonstrating that de novo designed protein binders can be functionally screened at scale inside living cells to discover proximity handles that induce degradation, stabilization, or organelle reorganization. The main caveats are context dependence, dropout bias, and limited dynamic range from a single primary reporter, but the orthogonal validations and endogenous functional readouts make the central claims robust Summary support

Interactive actions

Run deeper analyses or evolve this review with BGPT tools:

Selected primary citation

Multiplex proteome editing preprint