Recep Adiyaman
Vision About Experience Education Skills Publications Tools Newsletter Updates Podcasts Blog Contact CV Subscribe
All Issues
Older Newer
Daily Signal February 24, 2026 · 9 min read

Issue #54: Self-Aware Object Detection via Degradation Manifolds

Protein Design Digest - 2026-02-24 - Discovery of potent ALK tyrosine kinase inhibitors for thyroid cancer via machine learning modeling, molecular docking, MD simulations, and DFT study.

Share X LinkedIn
Protein Design Daily

Building something in Protein Design?

I love collaborating on new challenges. Let's build together.

LinkedIn Email

Subscribe to Protein Design Digest

Daily curated signals from arXiv, PubMed, and BioRxiv.

Signal of the Day

Self-Aware Object Detection via Degradation Manifolds

Object detectors achieve strong performance under nominal imaging conditions but can fail silently when exposed to blur, noise, compression, adverse weather, or resolution changes. In safety-critical settings, it is therefore insufficient to produce predictions without assessing whether the input remains within the detector’s nominal operating regime. We refer to this capability as self-aware object detection. We introduce a degradation-aware self-awareness framework based on degradation manifolds, which explicitly structure a detector’s feature space according to image degradation rather than semantic content. Our method augments a standard detection backbone with a lightweight embedding head trained via multi-layer contrastive learning. Images sharing the same degradation composition are pulled together, while differing degradation configurations are pushed apart, yielding a geometrically organized representation that captures degradation type and severity without requiring degradation labels or explicit density modeling. To anchor the learned geometry, we estimate a pristine prototype from clean training embeddings, defining a nominal operating point in representation space. Self-awareness emerges as geometric deviation from this reference, providing an intrinsic, image-level signal of degradation-induced shift that is independent of detection confidence. Extensive experiments on synthetic corruption benchmarks, cross-dataset zero-shot transfer, and natural weather-induced distribution shifts demonstrate strong pristine-degraded separability, consistent behavior across multiple detector architectures, and robust generalization under semantic shift. These results suggest that degradation-aware representation geometry provides a practical and detector-agnostic foundation.

Why this matters:

Also Worth Reading

Improving Protein Structure Prediction Using Integrative Cryo-EM and Ion Mobility Mass Spectrometry Modeling.

Proteins perform essential roles across nearly all cellular processes, and accurate three-dimensional structures remain critical for elucidating structure-function relationships and studies on drug discovery. Cryo-electron microscopy (cryo-EM), X-ray crystallography, and nuclear magnetic resonance can provide detailed structural information. However, for many proteins, structural information is available only as lower-resolution experimental data or sparse data. Such information is more difficult to translate into accurate atomic coordinates; a common example is low-resolution cryo-EM density maps. In parallel, mass spectrometry-based methods, including ion mobility (IM-MS), offer rapid, broadly applicable structural descriptors such as collisional cross section (CCS), a global measure of molecular shape and size, but CCS values also do not provide atomistic detail. Here we present CRIM (cryo-EM + IM-MS), an integrative Rosetta scoring function that combines low-resolution cryo-EM density information with IM-MS derived CCS as restraints to improve monomeric protein structure prediction. CRIM incorporates the Rosetta REF2015 (RS) energy with a CCS agreement penalty (computed via PARCS) and an electron-density agreement term (elec_dens_fast). We tested CRIM on an ideal dataset of 60 monomeric proteins using simulated CCS values and density maps. Across the ideal dataset, the CRIM score function improved or maintained prediction quality for many targets, reducing the mean RMSD from 3.65 Å (RS) to 2.90 Å and increasing the mean TM-score from 0.88 to 0.90. Furthermore, an experimental benchmark dataset of 54 proteins was curated to include either experimental cryo-EM maps or published CCS values. On the experimental dataset, CRIM similarly improved model selection, lowering the mean RMSD from 6.65 Å to 4.38 Å and raising the mean TM-score from 0.73 to 0.79. In comparison to AlphaFold3 predictions, CRIM frequently yielded competitive predictions and was able to substantially outperform AlphaFold3 for select difficult targets where sparse experimental restraints provide strong discriminatory power. The CRIM score function is freely available within the Rosetta software suite and provides a practical framework for leveraging complementary IM-MS and cryo-EM data to improve monomeric protein structure prediction.

Development of DHODH inhibitors incorporating virtual screening, pharmacophore modeling, fragment-based optimization methods, ADMET, molecular docking, molecular dynamics, PCA analysis, and free energy landscape.

The overexpression of dihydroorotate dehydrogenase (DHODH) in various malignant tumor cells is significantly associated with ferroptosis, making DHODH inhibition a promising strategy for cancer therapy. In this study, we employed an integrated approach to screen and optimize DHODH inhibitor candidates. First, virtual screening of the FDA-approved drug library identified 20 potential compounds (with the positive control AG-636 as a benchmark, docking score: 133.166). Subsequent pharmacophore modeling (ROC curve value >0.8) further narrowed the candidates to six compounds, which underwent fragment displacement optimization. All optimized compounds were evaluated for absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties. Molecular docking identified compounds 65:[(4S)-2,2-dimethyl-1,3-dioxolan-4-yl]methyl 3-(4-{[(2S)-2-hydroxypropyl]oxy}phenyl) (docking score: 197.362) and 66: [(4S)-2,2-dimethyl-1,3-dioxolan-4-yl]methyl 4-(4-{[(2S)-2-hydroxypropyl]oxy}phenyl) (docking score: 202.623) as high-affinity candidates. Molecular dynamics (MD) simulations, principal component analysis (PCA), and free energy landscape (FEL) analyses confirmed stable binding conformations for both compounds. Notably, compound 66: [(4S)-2,2-dimethyl-1,3-dioxolan-4-yl]methyl 4-(4-{[(2S)-2-hydroxypropyl]oxy}phenyl) exhibited minimal conformational changes, suggesting superior binding stability. This study advances compound 66: [(4S)-2,2-dimethyl-1,3-dioxolan-4-yl]methyl 4-(4-{[(2S)-2-hydroxypropyl]oxy}phenyl) as a promising DHODH inhibitor candidate through a multimodal workflow integrating structure-based pharmacophore design, fragment optimization, ADMET profiling, and advanced molecular simulations, providing a novel avenue for DHODH-targeted antitumor therapies.

Protein Engineering and Drug Discovery: Importance, Methodologies, Challenges, and Prospects.

Protein engineering is a rapidly evolving field that plays a critical role in transforming drug discovery and development. This innovative field harnesses the unique structural and functional properties of engineered proteins, such as monoclonal antibodies, nanobodies, therapeutic enzymes, and cytokines, to address complex diseases more effectively than traditional small-molecule drugs. These biologics not only enhance therapeutic specificity but also minimize adverse effects, marking a significant advancement in patient care. However, the journey of protein engineering is not without challenges. Issues related to protein folding, stability, and potential immunogenicity pose significant complications. Additionally, navigating the complex regulatory landscape can delay the transition from laboratory to clinical application. Addressing these hurdles requires the integration of cutting-edge technologies, including phage and yeast display technology, CRISPR, and advanced computational modeling, which enhance the predictability and efficiency of protein design. In this review, we explore the multifaceted impact of protein engineering on modern medicine, highlighting its potential to transform treatment paradigms, methodologies, challenges, and the successful development and approval of recombinant protein-based therapies. By navigating the complexities and leveraging technological advancements, the field is poised to unlock new therapeutic possibilities, ultimately improving patient outcomes and transforming healthcare.

Research & AI Updates

From the Industry

Quick Reads

From sweetener to risk factor: Network toxicology, molecular docking and molecular dynamics reveal the mechanism of aspartame in promoting coronary heart disease.

Aspartame, a widely used non-nutritive sweetener, has been epidemiologically linked to coronary heart disease (CHD), although the underlying mechanisms remain unclear. Read more →

Diethyl Phthalate (DEP) as a potential osteosarcoma risk factor: a multi-omics study integrating network Toxicology, single-cell RNA sequencing, and molecular docking.

Diethyl phthalate (DEP), a common plasticiser and endocrine disruptor, has been linked to cancer, but its role in osteosarcoma (OS) remains unclear. Read more →

Transition Metal Complex Based On Phenoliccarboxylic Ligand: Spectroscopic Inspection, DFT/TDDFT Computations, Cytotoxicity and Molecular Docking Investigation.

The study was designed to explore the synthesis, characterization, and biological evaluation of a novel Mn-(II)-3-[(3-carboxy-2-hydroxyphenyl)-methyl]-2-hydroxybenzoic complex, [Mn-(CHB-2H)·4H2O)]·2H2O, in comparison to its corresponding ligand (CHB). Read more →

Corrigendum to “Comparison on inhibitory effect and mechanism of inhibitors on sPPO and mPPO purified from ‘Lijiang snow’ peach by combining multispectroscopic analysis, molecular docking and molecular dynamics simulation” [Food Chem. 400 (2023) 134048].

Deciphering the potential of flavonols as SARS-CoV-2 MPro inhibitors: an in silico investigation using pass prediction, molecular docking, molecular dynamics simulation and ADMET analysis.

In the fight against contagious diseases, COrona VIrus disease 2019 (COVID -19) has been a formidable oponent. Read more →

Self-Aware Object Detection via Degradation Manifolds

Object detectors achieve strong performance under nominal imaging conditions but can fail silently when exposed to blur, noise, compression, adverse weather, or resolution changes. Read more →

Calibrated Adaptation: Bayesian Stiefel Manifold Priors for Reliable Parameter-Efficient Fine-Tuning

Parameter-efficient fine-tuning methods such as LoRA enable practical adaptation of large language models but provide no principled uncertainty estimates, leading to poorly calibrated predictions and unreliable behavior under domain shift. Read more →

Retraction: Molecular docking and dynamics simulation study of bioactive compounds from Ficus carica L. with important anticancer drug targets.

Pipeline Tip

Use local MSA generation (colabfold_search) to bypass speed bottlenecks.

Resources & Tools

Deep learning is not a magic wand, but a powerful lens for structural biology. — Recep Adiyaman

BS HF DK