The review was conducted using a PRISMA-guided systematic literature review methodology, following the Kitchenham guidelines for reproducible evidence synthesis. The authors searched three academic databases, Web of Science, PubMed, and IEEE Xplore, on 2 March 2026, using structured Boolean queries built around large language model terms, protein language model terms such as transformer, BERT, and ESM, and protein structure prediction concepts. The initial search identified 284 records; after deduplication, title and abstract screening, and full-text evaluation against predefined inclusion criteria, the final corpus settled at 38 studies spanning 2020 to 2025. Two reviewers conducted screening with adjudication by a third author, and the complete data extraction sheet has been released as supplementary material to allow independent verification of the entire process.

At the heart of the review lies a technical distinction that is often blurred in popular coverage. Protein language models, or PLMs, are transformer-based models pretrained on protein sequence or structure-augmented sequence data, exemplified by systems such as ESM, ProtBERT, and ProtTrans. General-purpose large language models, by contrast, are trained primarily on natural-language corpora and appear in this literature only when adapted to biological tasks such as literature mining or annotation. The review’s corpus is dominated by dedicated PLMs, which account for 19 of the 38 studies, or 50 percent, followed by general-purpose transformer adaptations at 18.4 percent, with the remainder comprising review and perspective articles. This concentration signals that domain-specific sequence representation learning has become the field’s dominant paradigm.

The technical machinery behind these models is elegant. Transformers employ self-attention mechanisms that capture long-range dependencies within sequential data, and when amino acid sequences are treated as tokens, the models learn contextual relationships between residues directly from massive protein sequence databases. The resulting embeddings encode latent biochemical, evolutionary, and structural information, capturing evolutionary conservation patterns and functional motifs without explicit supervision. Yet the review is careful to note a crucial caveat: sequence-only transformers do not explicitly encode the physicochemical and geometric constraints that govern protein folding, and high-dimensional embedding similarity should not be overinterpreted as evidence of biological realism, particularly for dynamic or poorly characterised protein systems.

To evaluate the field systematically, the authors propose a Biological-Readiness Taxonomy organised around six complementary dimensions: the source of biological knowledge, the representation mechanism, structural reasoning capacity, the biological prediction tasks addressed, validation maturity, and interpretability. Rather than classifying studies by model family, the taxonomy asks whether a model can support biologically meaningful structural inference. The framework is operationalised as an explicit, reproducible scoring rubric, with each dimension scored on an ordinal scale and a composite biological readiness score reported for every study in the corpus. The authors draw an analogy to Technology Readiness Level frameworks used in applied science, and to arguments that AlphaFold predictions constitute testable hypotheses rather than validated structures, but their instrument is domain-specific and multi-dimensional rather than generic.

Applying this framework reveals a clear evolutionary trajectory. The literature has shifted decisively from sequence-only representation learning toward hybrid and multimodal frameworks that integrate evolutionary information from multiple sequence alignments, structural priors, graph-based reasoning, and other biological signals. The review’s quantitative comparison of nine representative foundation-scale models, including AlphaFold2, RoseTTAFold, ESMFold, OmegaFold, AlphaFold-Multimer, AlphaFold3, RoseTTAFold All-Atom, Chai-1, and Boltz-1, traces a progression from single-chain tertiary structure prediction toward comprehensive biomolecular interaction modelling encompassing protein complexes, nucleic acids, and ligands. Earlier methods predicted the shape of individual proteins; the newest systems attempt to model the entire molecular choreography of the cell.

But the validation maturity analysis delivers the review’s most striking numbers. Among the nine representative models, benchmark evaluation remains the predominant validation strategy, accounting for 88.9 percent, or eight of nine models. Only one, AlphaFold3, extended evaluation beyond conventional benchmarking by explicitly comparing predictions against experimentally determined structures. None of the nine reported independent external dataset validation or prospective wet-laboratory validation. The authors are careful to frame this as an observation within an illustrative subset rather than a statistically representative estimate for the entire field, but the pattern is consistent with the corpus-wide taxonomy scores: research priorities remain heavily weighted toward benchmark accuracy, embedding quality, and computational performance, while interpretability, mechanistic biological reasoning, and experimental verification receive comparatively little attention.

The review identifies several systemic challenges that recur across many independent studies. Benchmark dependence is prominent: standardised datasets such as CASP and the Protein Data Bank enable reproducible comparison but draw on well-characterised protein families with extensive homologous sequence information, potentially yielding overly optimistic assessments of robustness on structurally heterogeneous proteins, intrinsically disordered regions, and dynamic conformational states. Transformer architectures also carry quadratic computational complexity with respect to sequence length, limiting scalability to large proteins and proteome-wide applications. Interpretability remains perhaps the most persistent weakness: most models achieve high predictive performance while the reasoning behind their predictions stays opaque, and the few explainability efforts that exist typically offer model-level information such as attention weights rather than biologically valid, experimentally testable explanations.

These limitations matter far beyond academic benchmarking. Accurate protein structure prediction underpins drug discovery, therapeutic protein engineering, biomarker identification, disease diagnosis, and personalised medicine, and the review’s clinical discussion makes clear that the documented gaps in biological interpretability, external validation, and experimental verification would need to close before language-model-based structural predictions could be routinely trusted in therapeutic contexts. The authors also acknowledge the limitations of their own work with unusual candour: the search was restricted to English-language studies in three databases, taxonomy scoring was performed by a single author without inter-rater reliability statistics, and the taxonomy itself remains an unvalidated instrument intended as a starting point for discussion rather than a settled standard.

The review’s ultimate message is a call to rebalance the field’s priorities. Future progress, the authors argue, should be evaluated not only by benchmark accuracy but also by biological validation, structural reasoning capability, interpretability, and translational applicability. They point toward biologically grounded modelling frameworks that integrate structural constraints, evolutionary information, physicochemical knowledge, and multimodal evidence, alongside continued advances in computational efficiency, lightweight foundation models, and explainable AI. The Biological-Readiness Taxonomy offers a practical instrument for tracking that progress, giving researchers and funders a structured way to distinguish models that merely top the leaderboards from those genuinely ready to accelerate discovery in the laboratory. For a field that has reshaped structural biology in half a decade, the next frontier may not be predicting structures better, but proving that the predictions can be believed.

Subject of Research: Systematic review of protein language models for protein structure prediction and a proposed biological readiness taxonomy

Article Title: A systematic review of protein language models for protein structure prediction and a proposed biological readiness taxonomy

Article References: Al-Quraishi, T., Zaboon, W. A., Abawajy, J., & Nowrozy, R. (2026). A systematic review of protein language models for protein structure prediction and a proposed biological readiness taxonomy. Discover Artificial Intelligence, 6(1), Article 1338. https://doi.org/10.1007/s44163-026-02333-6

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02333-6

Keywords: protein language models, protein structure prediction, AlphaFold, transformers, biological validation, interpretability, systematic review, benchmarking, multimodal learning, drug discovery, computational biology, taxonomy

The research was born of necessity. Temminck’s pangolin (Smutsia temminckii) is classified as Vulnerable by the International Union for Conservation of Nature, and its populations have declined drastically across its range due to poaching for illegal wildlife and traditional medicine trades, habitat loss, electric fences, and overexploitation. Individuals rescued from traffickers often arrive at rehabilitation facilities in poor condition and require extended care before they can be released. Yet the species is notoriously stress-prone, and ethical, logistical, and conservation constraints make it impossible to capture, hold, or conduct experimental manipulations on pangolins purely for research purposes. Traditional physiological validation, which involves administering a compound such as adrenocorticotropic hormone to artificially activate the stress axis, was therefore off the table.

To understand why this matters, it helps to look at how stress hormones work. The physiological stress response involves two main systems: the sympathetic nervous system, which triggers the familiar ‘fight or flight’ release of catecholamines, and the hypothalamic–pituitary–adrenal (HPA) axis, which drives the secretion of glucocorticoid steroid hormones. An acute rise in glucocorticoids is adaptive, mobilising energy, enhancing cardiovascular activity, and shifting behaviour to restore homeostasis. But chronic elevation carries serious costs, including immune and reproductive suppression and a general decline in fitness. Researchers therefore track glucocorticoid concentrations as a proxy for the stress an individual experiences and its likely consequences for survival and reproduction.

Blood sampling, the traditional route for hormone monitoring, is particularly problematic for pangolins. Capturing and restraining an animal to draw blood activates the very stress response researchers want to measure, with blood glucocorticoids rising significantly within three to five minutes of capture. Repeated collections are impractical for small-bodied species with limited blood volumes, and serum or plasma hormone levels fluctuate with pulsatile secretion patterns, circadian rhythms, and feeding state. For threatened species, invasive endocrine monitoring has fallen out of favour across much of the research community.

Non-invasive monitoring sidesteps these problems. Once steroid hormones are secreted into the bloodstream by the HPA axis, the liver metabolises them and excretes them in bile and faeces. Because metabolites pool in the gut, faecal glucocorticoid metabolite (fGCM) concentrations provide a smoother, longer-term signal spanning several hours, rather than a snapshot distorted by episodic hormone pulses. Collecting faeces requires no direct human–animal interaction, eliminates stress-related feedback from handling, and allows repeated longitudinal sampling from small-bodied species. But there is a catch: hormone metabolism and excretion are sex- and species-specific, so any EIA must be validated in each species before its results can be trusted.

The team, led by Juan Scheun of the BioPulse Research Group at Tshwane University of Technology, together with colleagues from the South African National Biodiversity Institute, Tshwane University of Technology, and the Mammal Research Institute at the University of Pretoria, turned to biological validation. They opportunistically collected 28 faecal samples from five Temminck’s pangolins—three females and two males—confiscated from the illegal wildlife trade and admitted to the Johannesburg Wildlife Veterinary Hospital during 2020 and 2021. Samples were gathered before and after known stressors such as confiscation, veterinary handling, and transport, with samples collected during periods of no handling serving as baselines. For security reasons, the animals were housed at an undisclosed off-site location, and the hospital’s welfare policies meant no strict sampling regime could be imposed; staff simply collected fresh faeces whenever animals defecated, sometimes days apart.

In the laboratory, frozen samples were lyophilised, pulverised, and sieved, then extracted with 80% ethanol and analysed at the Endocrine Research Laboratory of the University of Pretoria using five different EIAs: a cortisol assay, a corticosterone assay, two 11-oxoaetiocholanolone assays (designated 72a and 72T), and a 5α-pregnane-3ß,11ß,21-triol-20-one assay (37e). Serial dilutions produced displacement curves parallel to the standard curves in all assays, confirming analytical validity. The researchers then applied a stringent criterion: an assay was considered reliable only if it detected peak fGCM increases exceeding 150% above baseline, a tougher threshold than the 100% often used in comparable studies, to reduce the risk of false positives.

The results were striking. In female 1, the 72a assay recorded a 982% increase in fGCM concentrations following a stressful event, with 72T close behind at 763%, followed by the cortisol assay at 481% and 37e at 317%. Female 2 showed increases above the 150% threshold for 37e (325%), 72T (221%), and 72a (201%), and her secondary fGCM rise coincided with a move to a holding facility ahead of release. Female 3 exceeded the threshold on both 72a (339%) and 72T (251%). Among the males, only 72a detected increases above 150% in both individuals, reaching 168% in male 1 and 695% in male 2, where the cortisol assay also responded strongly at 437%. Notably, in female 2, fGCM levels rose considerably before a decline in health that led to her readmission to the rehabilitation centre, hinting that hormone monitoring might one day flag deteriorating welfare before clinical signs appear.

The standout performer was the 72a assay, the only one to exceed the 150% threshold in every study animal of both sexes, making it the recommended tool for monitoring fGCM levels in Temminck’s pangolin. The 72T assay proved a strong alternative for females, while 37e, which exceeded 100% increases in all females, can serve as a back-up when the primary assays are unavailable. The cortisol assay should be treated as equivocal pending further validation. Interestingly, the finding that cortisol metabolites dominated the faecal profile contrasts with work on the Taiwanese pangolin, where mass spectrometry revealed higher faecal corticosterone concentrations, though that study measured actual hormones rather than immunoreactive metabolites, and the authors suggest mass spectrometry could further clarify active circulating glucocorticoids in Temminck’s pangolin.

Beyond the assay validation itself, the study delivers an uncomfortable but important message: direct human interaction, even in the context of care and rehabilitation, elicits pronounced physiological stress responses in pangolins, with increases exceeding 100% across all study animals. Similar patterns have been documented in rehabilitated African penguins and koalas. The findings argue for refining rehabilitation protocols to minimise human contact and for assessing individuals physiologically before release to improve survival odds. The authors caution that the small sample size limits generalisability, that excretion time lags could not be determined for the species—gut passage times in other pangolins range from roughly 20 to 60 hours—and that individual variation in stress perception likely shapes adrenal responses. Even so, the validated 72a EIA now offers researchers, conservationists, and rehabilitation managers a way to track stress in captive settings and, crucially, in free-ranging populations, where pressures such as climate change are expected to intensify. For an animal that rolls into an armour-plated ball when threatened, the ability to read its stress from a dropping may prove one of the most powerful conservation tools yet devised.

Subject of Research: Biological validation of non-invasive faecal glucocorticoid metabolite enzyme immunoassays for measuring physiological stress in rehabilitated Temminck's pangolins

Article Title: Biological validation of enzyme immunoassays for measuring physiological stress in rehabilitated Temminck’s pangolins

Article References: Scheun, J., Labuschagne, K., Jansen, R., Ganswindt, A., & Long, C. (2026). Biological validation of enzyme immunoassays for measuring physiological stress in rehabilitated Temminck’s pangolins. Discover Conservation, 3(1), Article 20. https://doi.org/10.1007/s44353-026-00090-2

Image Credits: AI Generated

DOI: 10.1007/s44353-026-00090-2

Keywords: Temminck's pangolin, faecal glucocorticoid metabolites, enzyme immunoassay, stress physiology, wildlife rehabilitation, non-invasive hormone monitoring, conservation, HPA axis, animal welfare, illegal wildlife trade, biological validation, South Africa