CoChat
Sign Up
Cited by 50Protein Degradation and InhibitorsComputational Drug Discovery MethodsChemical Synthesis and Analysis

Advancing Targeted Protein Degradation via Multiomics Profiling and Artificial Intelligence

Open access
Duran-Frigola, Miquel, Cigler, Marko, Winter, Georg E.·2023·Journal of the American Chemical Society
DOI 10.1021/jacs.2c11098PMC PMC9912273
Open this paper in CoChat — free account

Abstract

Only around 20% of the human proteome is considered to be druggable with small-molecule antagonists. This leaves some of the most compelling therapeutic targets outside the reach of ligand discovery. The concept of targeted protein degradation (TPD) promises to overcome some of these limitations. In brief, TPD is dependent on small molecules that induce the proximity between a protein of interest (POI) and an E3 ubiquitin ligase, causing ubiquitination and degradation of the POI. In this perspective, we want to reflect on current challenges in the field, and discuss how advances in multiomics profiling, artificial intelligence, and machine learning (AI/ML) will be vital in overcoming them. The presented roadmap is discussed in the context of small-molecule degraders but is equally applicable for other emerging proximity-inducing modalities.

AI snapshot

Written from the paper’s full text.

TL;DR

This perspective argues that targeted protein degradation (TPD) can overcome limits of occupancy-driven inhibitors, but progress is constrained by few usable E3 ligands, serendipitous molecular glue discovery, and difficulty predicting compatible POI–ligase pairs. It proposes a roadmap integrating multiomics profiling, phenotypic screening, and AI/ML for ligand discovery, degrader design, and target identification.

Key findings

  • Only around 20% of the human proteome is considered druggable with small-molecule antagonists, leaving targets such as transcription factors, intrinsically disordered proteins, and scaffolding proteins outside conventional inhibitor discovery.
  • TPD degraders can extend the druggable space and improve selectivity: PROTACs can convert ineffectual binders into functional degraders, and molecular glue degraders can induce degradation of proteins considered unligandable, such as IKZF1/IKZF3, RBM39, and cyclin K.
  • The paper identifies three central TPD challenges: expanding PROTAC reach through advanced ligand discovery, identifying MGDs via phenotypic screens with cutting-edge target ID, and predicting compatible POI–ligase pairs.
  • AI/ML and multiomics are proposed as key enablers, including chemoproteomics for native-state ligand discovery, AlphaFold-based structural coverage expansion from 48% to 76% of human proteome residues, and generative models for linker and degrader design.

Methods

  • Perspective/review article; no primary experimental data, sample, or new wet-lab results are reported.
  • Synthesizes literature across TPD, chemoproteomics, phenotypic screening, target identification, protein–protein docking, and AI/ML methods.
  • Main analysis is conceptual: it reviews existing strategies and proposes future computational and multiomics pipelines for degrader discovery.

Limitations

  • The paper states that there is currently no well-established computational framework to predict and evaluate compound-induced protein–protein interactions for MGD discovery.
  • The paper states that the small number of experimentally resolved ternary complexes complicates benchmarking of computational TPD modeling methods.
  • The paper states that AI/ML models often act as “black boxes,” lack confidence intervals, and have seen relatively slow adoption in drug discovery; it also describes de novo MGD design as “arguably rather a wish list.” Evident from the text: as a perspective, it provides no experimental validation of the proposed roadmap.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1021/jacs.2c11098

Cites

290 of 290 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 159 cited sources checked · 0 supported · 0 flagged

  • Findingcitation

    After AlphaFold2, structural coverage of the human proteome increased from 48% to 76% of total residues.

    “after AlphaFold2, the structural coverage of the human proteome increased from 48% to 76% of the total residues).131”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [131] source

      Not checked in the free run.

  • Findingcitation

    A 100-fold acceleration of docking was observed by docking a fraction of the chemical library in synchrony with an AI/ML ligand-based prediction of the remaining docking scores.

    “For example, a 100-fold acceleration of the process was observed by docking a fraction of the chemical library in synchrony with an AI/ML ligand-based prediction of the remaining docking scores.144”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [144] source

      Not checked in the free run.

  • Premisereasoning

    Even a protein classified as ligandable can pose significant challenges to drug discovery.

    “Upon closer inspection, even a protein that is classified as ligandable can pose significant challenges to the drug discovery process.”
    Quote in the paperSubstantial claim
  • Interpretationreasoning

    This leaves some of the most compelling therapeutic targets outside the reach of ligand discovery.

    “This leaves some of the most compelling therapeutic targets outside the reach of ligand discovery.”
    Quote in the paperSubstantial claim
  • Premisereasoning

    Even if a protein target can be meaningfully blocked by a small-molecule inhibitor, a potential therapeutic window can be compromised by toxicity from target inhibition in other tissues or cell types.

    “Lastly, even if a protein target can be meaningfully blocked by a small-molecule inhibitor, a potential therapeutic window can be compromised by toxicity that emerges from target inhibition in other tissues or cell types.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 43
  • May combine two claims. 23

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    TPD promises to overcome some of these limitations.

    “The concept of targeted protein degradation (TPD) promises to overcome some of these limitations.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Typically, degraders function by inducing proximity between a POI and an E3 ubiquitin ligase.

    “Typically, they function by inducing the proximity between a target protein of interest (POI) and an E3 ubiquitin ligase.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    The targeted bromodomain proteins also harbored multiple other functional domains and often participate in protein-protein interactions that appeared bromodomain independent.

    “In all cases, the targeted proteins also harbored multiple other functional domains and are often participating in intricate protein–protein interactions that appeared to be bromodomain independent.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    TPD is dependent on small molecules that induce proximity between a POI and an E3 ubiquitin ligase, causing ubiquitination and degradation of the POI.

    “TPD is dependent on small molecules that induce the proximity between a protein of interest (POI) and an E3 ubiquitin ligase, causing ubiquitination and degradation of the POI.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    HER3 features very low catalytic activity but engages in oncogenic signaling by acting as a scaffold to enable receptor heterodimerization.

    “HER3 features very low catalytic activity but engages in oncogenic signaling by acting as a scaffold to enable receptor heterodimerization.8−10”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [8] source

      Not checked in the free run.

    • cites [9] source

      Not checked in the free run.

    • cites [10] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    The advent of AI/ML technologies, coupled with multiomics profiling techniques for phenotypic screening, may pave the way toward efficient identification and rational design of small-molecule degraders.

    “In particular, we argue that the advent of AI/ML technologies, coupled with multiomics profiling techniques for phenotypic screening, may pave the way toward efficient identification and rational design of small-molecule degraders.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    PROTACs leverage two different ligands to simultaneously engage the POI and the ligase.

    “On the one hand, heterobifunctional proteolysis-targeting chimeras (PROTACs) leverage two different ligands to simultaneously engage the POI and the ligase.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Several mechanisms contribute to increased selectivity of PROTACs, including steric incapability of some targets to form a productive ternary complex, or lack of a suitably positioned lysine residue for ubiquitination.

    “Several mechanisms have been shown to contribute to this increase in selectivity. This includes steric incapability of some targets to form a productive ternary complex, or lack of a suitably positioned lysine residue for ubiquitination.62−69”
    Quote in the paperSubstantial claim0 of 8 sources checked
    • cites [62] source

      Not checked in the free run.

    • cites [63] source

      Not checked in the free run.

    • cites [64] source

      Not checked in the free run.

    • cites [65] source

      Not checked in the free run.

    • cites [66] source

      Not checked in the free run.

    • cites [67] source

      Not checked in the free run.

    • cites [68] source

      Not checked in the free run.

    • cites [69] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    PROTACs are in principle able to address proteins that can be liganded, but fall short for proteins considered unligandable, such as the majority of transcription factors.

    “PROTACs are therefore, in principle, able to address proteins that can be liganded, but fall short for proteins that are considered unligandable, such as the majority of transcription factors.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    MGDs connect ligase and POI by modulating protein surface topologies and orchestrating multiple PPIs.

    “On the other hand, molecular glue degraders (MGDs) connect ligase and POI by modulating protein surface topologies and orchestrating multiple PPIs between the involved proteins.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    Chemoproteomics aims to map ligand-protein interactions on a proteome-wide scale, typically in intact cells.

    “In contrast, chemoproteomics aims to map ligand–protein interaction on a proteome-wide scale, typically in intact cells.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Mature software tools now exist to run generative models on conventional computers, and community-accepted benchmarks enable systematic assessment of generated small molecules considering predicted activity, novelty, validity, and diversity.

    “Mature software tools now exist to run generative models on conventional computers,154,157 and community-accepted benchmarks enable the systematic assessment of generated small-molecules, considering not only their predicted activity but also their novelty, validity, and diversity.158,159”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [154] source

      Not checked in the free run.

    • cites [157] source

      Not checked in the free run.

    • cites [158] source

      Not checked in the free run.

    • cites [159] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A synthesis-on-a-chip procedure was devised in which an AI/ML tool suggested compounds that could be immediately synthesized by constraining compounds with a subset of simple reactions and building blocks.

    “For example, by constraining compounds with a subset of simple reactions and building blocks, a “synthesis-on-a-chip” procedure was devised where the AI/ML tool suggested compounds that could be immediately synthesized.161”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [161] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Existing platforms have been versioned to design optimal linkers connecting the two molecular subunits in a PROTAC.

    “Moreover, existing platforms have been versioned to design optimal linkers connecting the two molecular subunits in a PROTAC.163”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [163] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    A fundamental limitation of supervised AI/ML is the requirement of previously available experimental data.

    “A fundamental limitation of supervised AI/ML is the requirement of previously available experimental data.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Generative models have been used to compile focused and diversity-oriented chemical libraries, design inhibitors de novo, and perform hit-to-lead optimization, typically in a multiobjective mode.

    “Generative models have been used to compile both focused and diversity-oriented chemical libraries,149,150 to design inhibitors de novo,151 and to perform hit-to-lead optimization,152 typically in a multiobjective mode where e.g. selectivity, synthetic accessibility, and drug-like properties are improved simultaneously.153,154”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [149] source

      Not checked in the free run.

    • cites [150] source

      Not checked in the free run.

    • cites [151] source

      Not checked in the free run.

    • cites [152] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Proteomics-driven ligand discovery has the potential to significantly expand the reach of TPD and other proximity-inducing modalities.

    “In sum, proteomics-driven ligand discovery has the potential to significantly expand the reach of TPD and other proximity-inducing modalities.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The catalytic mechanism of degraders can potentially omit the need for continuously high drug levels, providing another opportunity to limit toxicity.

    “In addition, the catalytic mechanism of degraders can potentially omit the need for continuously high drug levels, thus providing another opportunity to limit toxicity.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Advanced AI/ML methods are generally data-greedy, requiring more data than classical QSAR techniques.

    “This is aggravated by the fact that advanced AI/ML methods are generally data-greedy, meaning they require more data than the classical set of QSAR techniques.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    Computational approaches to ligand discovery can be divided into two broad categories: supervised methods requiring pre-existing activity data and methods not requiring previous knowledge of binding affinity.

    “Computational approaches to ligand discovery can be divided into two broad categories (Figure 2).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    This enabled design of RNF114-based degraders, expanding to other chemotypes.

    “This enabled the design of RNF114-based degraders, also expanding to other chemotypes.84,85”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [84] source

      Not checked in the free run.

    • cites [85] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    The paucity of ligands for E3 ligases and currently undrugged proteins critically hampers progress in TPD.

    “The paucity of ligands for E3 ligases and currently undrugged proteins critically hampers progress in TPD.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Thermal proteomics profiling (TPP) has been employed to identify FECH as a frequent off-target of kinase inhibitors and leucine aminopeptidase 3 as an off-target of several HDAC inhibitors.

    “Thermal stabilization (i) is leveraged in thermal proteomics profiling (TPP), an approach that has been employed to, for instance, identify the ferrochelatase FECH as a frequent off-target of kinase inhibitors,224 or to uncover leucine aminopeptidase 3 as an off-target of several HDAC inhibitors.225”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [224] source

      Not checked in the free run.

    • cites [225] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    Converting raw -omics and molecular data into an AI/ML-friendly form is challenging due to high dimensionality, biases, artifacts specific to the -omics technique, and inconsistencies in units and identifiers.

    “Converting raw -omics and molecular data into an AI/ML-friendly form is a challenging task due to the high dimensionality of the data sets, biases, and artifacts specific to the -omics technique, and inconsistencies in units and identifiers.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Phenotypic screens are increasingly delivering richer outputs, including high-content microscopy readouts and postperturbation transcriptomics measures such as bulk- or single-cell RNA-seq.

    “Phenotypic screens are increasingly delivering richer outputs, including high-content microscopy readouts and postperturbation transcriptomics measures such as bulk- or single-cell RNA-seq (Figure 4).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    CRBN was identified as the protein target of thalidomide by a chemoproteomics strategy that employed a tethered thalidomide analogue.

    “A prototypic example is the identification of CRBN as the protein target of thalidomide, which was enabled by a chemoproteomics strategy that employed a tethered thalidomide analogue.48”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [48] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The LINCS consortium, and in particular the L1000 Connectivity Map resource, provides a catalog of postperturbation transcriptomics profiles for thousands of compounds, offering a global map of the phenotypic response landscape of the medicinal chemistry space.

    “Of note, the LINCS consortium, and in particular the L1000 Connectivity Map resource, provides a catalog of postperturbation transcriptomics profiles for thousands of compounds, offering a global map of the phenotypic response landscape of the medicinal chemistry space.205”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [205] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    Chemoproteomics approaches are the gold-standard to identify protein target(s) bound by a given small-molecule hit of interest.

    “Aforementioned chemoproteomics approaches are the gold-standard to identify protein target(s) bound by a given small-molecule hit of interest.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Recent genome-wide CRISPR screens with a single cell RNA-seq readout are providing vast collections of transcriptional phenotypes, with a clearly identifiable protein degradation cluster that can be used as a reference for small-molecule degrader discovery.

    “Recent genome-wide CRISPR screens with a single cell RNA-seq readout are providing vast collections of transcriptional phenotypes,211 with a clearly identifiable protein degradation cluster that can be used as a reference for small-molecule degrader discovery.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [211] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    By analyzing growth inhibition profiles across cell lines, it was found that the drug efflux pump MDR1 confers resistance to PROTACs.

    “By analyzing growth inhibition profiles across cell lines, it was for instance found that the drug efflux pump MDR1 confers resistance to PROTACs.202”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [202] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The computational pipeline PRosettaC starts by modeling POI-warhead and E3-recruitment ligand interactions, and then exploits PROTAC linker length constraints to drastically reduce the docking search space.

    “Similarly, the computational pipeline PRosettaC starts by modeling POI-warhead and E3-recruitment ligand interactions, and then exploits the fact that PROTAC linkers have length constraints to drastically reduce the docking search space (Figure 5).71”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [71] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The current state-of-the-art in sequence-based modeling is the use of AI/ML embeddings pretrained on multimillion-scale protein collections.

    “The current state-of-the-art in sequence-based modeling is the use of AI/ML embeddings pretrained on multimillion-scale protein collections.118”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [118] source

      Not checked in the free run.

  • May combine two claims.

    Premisereasoning

    The ligand plays an important role in determining the PPI interface, either by reshaping it to increase affinity (as in every MGD) or by imposing occupancy and distance constraints (as in PROTACs).

    “While structural modeling with apo (unbound) forms of the POI and E3 monomers may be sufficient in some ternary PROTAC complexes, it is generally recognized that the ligand plays an important role in determining the PPI interface, either by reshaping it to increase affinity (as it happens in every MGD) or by imposing occupancy and distance constraints (as in the case of PROTACs).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Methods to assess the PROTACtability of POIs have been suggested, prioritizing over a thousand proteins with potential for becoming PROTAC targets.

    “For example, methods to assess the PROTACtability of POIs have been suggested,245 prioritizing over a thousand proteins with potential for becoming PROTAC targets.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [245] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The holo form of CRBN was used in a large-scale docking run to predict C2H2 zinc-finger degrons.

    “Indeed, the holo form of CRBN was used in a large-scale docking run to predict C2H2 zinc-finger degrons.51”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [51] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Systematic screening efforts have revealed high glue hit rates in drug-like collections, suggesting that glue-like traits may be more abundant than previously anticipated.

    “In addition, systematic screening efforts have revealed high glue hit rates in drug-like collections,59 suggesting that glue-like traits may be more abundant than previously anticipated.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [59] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    These embeddings can capture evolutionary traits of proteins, as well as functional and structural characteristics.

    “These embeddings can capture evolutionary traits of the proteins, as well as functional and structural characteristics.119”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [119] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Baseline compatibility between the POI and the E3 is a necessary requirement for the formation of a ternary complex.

    “It has been suggested that baseline compatibility between the POI and the E3 is a necessary requirement for the formation of a ternary complex.61,239”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [61] source

      Not checked in the free run.

    • cites [239] source

      Not checked in the free run.

  • May combine two claims.

    Premisenone given

    Extensive validation beyond canonical examples and benchmarking against molecular dynamics simulations for AlphaFold2 ensemble generation are still lacking.

    “although extensive validation beyond canonical examples and benchmarking against molecular dynamics simulations are still lacking.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Allosteric sites are more ligandable than PPI interfaces, in general.

    “Allostery has been suggested as a promising approach to modulating PPIs,261 since allosteric sites are more ligandable than PPI interfaces, in general.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [261] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Adoption of AI/ML methods in this field has been relatively slow and often received with skepticism, compared to other industrial applications.

    “Arguably, though, adoption of AI/ML methods in this field has been relatively slow and often received with skepticism, compared to other industrial applications.271”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [271] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Counterfactual analysis often reveals punctual modifications of the input that cause abrupt changes in the output.

    “This often reveals punctual modifications of the input that cause abrupt changes in the output (a.k.a. “counterfactuals”),276”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [276] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Databases of structural ensembles of proteins exist.

    “Moreover, databases of structural ensembles of proteins exist264”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [264] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The two-step allosteric MGD design approach is easy to parallelize.

    “While an approach like this one may be computationally intensive, it is easy to parallelize.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

    Premisecitation

    Incorporating kinetic models for ternary complex formation remains a long-term goal.

    “Despite its relevance to resolve complete and incomplete degradation behaviors, this remains a long-term goal.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [270] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationnone given

    As standalone tools, most AI/ML methods do not offer sufficient evidence to support their predictions.

    “One reason for this is the fact that, as standalone tools, most AI/ML methods do not offer sufficient evidence to support their predictions.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Most AI/ML methods are presented as “black boxes” that ingest an input and produce an output, without providing physical or mechanistic insights.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    "Explainable" AI/ML benchmarks are flourishing, aimed at finding substructures in a molecule that determine bioactivity, in agreement with medicinal chemistry expertise.

    “Accordingly, “explainable” AI/ML benchmarks are flourishing, aimed at finding substructures in a molecule that determine bioactivity, in agreement with medicinal chemistry expertise.273”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [273] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Initiatives like DREAM Challenges, MoleculeNet, and Therapeutic Data Commons offer benchmark data sets across a wide spectrum of prediction tasks for pharmacology.

    “Initiatives like the DREAM Challenges,284 MoleculeNet,110 and Therapeutic Data Commons285 offer benchmark data sets across a wide spectrum of prediction tasks for pharmacology.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [284] source

      Not checked in the free run.

    • cites [110] source

      Not checked in the free run.

    • cites [285] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    The aforementioned screening assays, as well as scalable methods to discover drug-induced PPIs such as Y2H-based/related assays, may generate the necessary data to include QSAR modeling in the loop of design.

    “In this line, the aforementioned screening assays, as well as scalable methods to discover drug-induced PPIs such as Y2H-based/related assays,268 may generate the necessary data to include QSAR modeling in the loop of design.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [268] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    AlphaFold2 provides a per-residue confidence measure that allows researchers to judge the quality of the predicted structure.

    “Likewise, AlphaFold2 provides a per-residue confidence measure that allows researchers to judge the quality of the predicted structure.281”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [281] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Making more transparent AI/ML models has been identified as a key factor to gain trust in AI/ML.

    “Making more transparent AI/ML models, for instance by highlighting the structural moieties that determine a binding event, has been identified as a key factor to gain trust in AI/ML.272”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [272] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Systematic annotation of putative allosteric binding cavities is possible and has already been attempted on a proteome-wide scale.

    “Systematic annotation of putative allosteric binding cavities is also possible, and has already been attempted on a proteome-wide scale.267”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [267] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    In allosteric MGD function, the small molecule does not participate directly in the interaction; rather, it exposes or modifies a PPI interface by binding to a distant pocket that stabilizes a (non-native) conformation of the protein.

    “In this case, the small molecule does not participate directly in the interaction. Rather, it exposes or modifies a PPI interface by binding to a distant pocket that stabilizes a (non-native) conformation of the protein.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [260] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Recent modeling attempts have been done to predict DC50 and Dmax concentrations of novel PROTACs.

    “Recent modeling attempts have been done to predict half maximal (DC50) and maximal degradation (Dmax) concentrations of novel PROTACs.269”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [269] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Proteins have an intrinsic, evolved tendency to form complexes, with random mutations easily prompting the emergence of new interfaces.

    “Likewise, it has been shown that proteins have an intrinsic, evolved tendency to form complexes,254 with random mutations easily prompting the emergence of new interfaces.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [254] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Supervised AI/ML methods are likely to become more relevant as more instances of small-molecule degraders are discovered.

    “Supervised AI/ML methods are likely to become more relevant as more instances of small-molecule degraders are discovered.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Modern algorithms such as transformer models have been applied to identify relevant atoms in a ligand or to pinpoint important residues in a protein binding site.

    “Modern algorithms such as “transformer models”, which have found tremendous success in the field of natural language processing and computer vision, have been applied to identify relevant atoms in a ligand or to pinpoint important residues in a protein binding site.274”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [274] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Statistical methods to quantify contribution of certain chemical features to the predicted outcome have been suggested, borrowing concepts from game theory.

    “Similarly, statistical methods to quantify contribution of certain chemical features to the predicted outcome have been suggested, borrowing concepts from “game theory”.275”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [275] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    MGDs may also function from an allosteric site.

    “Finally, MGDs may also function from an allosteric site.260”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [260] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The same confidence principle applies to transcriptomics profiles of cell lines.

    “The same applies to e.g. transcriptomics profiles of cell lines.279”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [279] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The majority of existing AI/ML models remain in specialized journals, with poorly maintained code and variable rigor in validation.

    “Unfortunately, though, the majority of existing AI/ML models remain in specialized journals, with poorly maintained code and variable rigor in validation.283”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [283] source

      Not checked in the free run.

61 of 61

  • The paper relies on this without support.

    The presented roadmap is equally applicable for other emerging proximity-inducing modalities.

    “The presented roadmap is discussed in the context of small-molecule degraders but is equally applicable for other emerging proximity-inducing modalities.”

  • The paper relies on this without support.

    Most ligand discovery efforts are focused on small molecules that bind to hydrophobic pockets and aim to block the biochemical activity encoded by the bound site.

    “Most ligand discovery efforts are focused on small molecules that bind to hydrophobic pockets and aim to block the biochemical activity encoded by the bound site.”

  • The paper relies on this without support.

    Computational methods to enable prospective and rational design of degraders are still scarce.

    “Overall, computational methods to enable prospective and rational design of degraders are still scarce.”

  • The paper relies on this without support.

    Phenotypic screens have a successful history in the identification of MGDs.

    “We have reflected on the successful history of phenotypic screens in the identification of MGDs”

  • The paper relies on this without support.

    Most AI/ML methods are presented as “black boxes” that ingest an input and produce an output, without providing physical or mechanistic insights.

    “Rather, they are presented as “black boxes” that ingest an input (e.g., a molecule structure) and produce an output (e.g., a binding affinity prediction), without providing physical or mechanistic insights.”

  • The paper relies on this without support.

    Current efforts are focused on identifying “causal” links between the input and the output of an AI/ML model.

    “In addition, current efforts are focused on identifying “causal” links between the input and the output of an AI/ML model.”

  • The paper relies on this without support.

    AI/ML models will be more confident if the input small molecule belongs to the same chemical space that was used at training time, while predictions will be less reliable if the input has completely new characteristics previously unseen by the model.

    “Naturally, AI/ML models will be more confident if the input small molecule belongs to the same chemical space that was used at training time, while predictions will be less reliable if the input has completely new characteristics previously unseen by the model.”

  • The paper relies on this without support.

    The rapid adoption of AlphaFold2 can only be explained by unprecedented postpublication efforts to make the technology accessible to the broad scientific community.

    “The rapid adoption of AlphaFold2 can only be explained by unprecedented postpublication efforts to make the technology accessible to the broad scientific community.”
8 of 8

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • AI impacts most therapeutic modalities, including targeted protein degradation, antibodies, gene therapy, oligonucleotide design, and vaccine design.

    “AI impacts most therapeutic modalities, including targeted protein degradation (11), antibodies (12), gene therapy (13), and oligonucleotide (14) and vaccine (15) design.”

    From Artificial Intelligence for Drug Discovery: Are We There Yet? (2024) · cites this paper as [12]

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Duran-Frigola, M., Cigler, M., & Winter, G. E.
Affiliations
  • CeMM Research Center for Molecular Medicine of the Austrian Academy of Sciences, 1090 Vienna, Austria
  • Austrian Academy of Sciences (AT)
Corresponding author
Winter, Georg E.
ORCID
Venue
Journal of the American Chemical Society
Publisher
American Chemical Society (ACS)
Published
2023-01-27
Volume
145
Issue
5
Pages
2711-2732
ISSN
  • 0002-7863 (ISSN-L)
  • 1520-5126
DOI
10.1021/jacs.2c11098
PMC
PMC9912273
OpenAlex
W4318257682
Semantic Scholar
ae174becc2ccd501a7967e7a403088423eb92b6e
Cited by
50
References
290
Citations by year
2026: 11 · 2025: 21 · 2024: 14 · 2023: 4
FWCI
1.32
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of the American Chemical Society (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • Austrian Science Fund (P31690, P32125, P7909)
  • H2020 European Research Council (851478)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Degradation and Inhibitors
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Computational Drug Discovery Methods
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Chemical Synthesis and Analysis
Keywords
targeted protein degradation, small-molecule degraders, E3 ubiquitin ligase, ubiquitination, protein degradation, druggable proteome, multi-omics profiling, artificial intelligence, machine learning, drug discovery
MeSH terms
Multiomics, Artificial Intelligence, Humans, Ubiquitin-Protein Ligases, Ubiquitination, Proteolysis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review