Recep Adiyaman
Vision About Experience Education Skills Publications Tools Newsletter Updates Podcasts Blog Contact CV Subscribe
All Issues
Older Newer
Weekly Digest February 27, 2026 · 22 min read

Weekly Digest: Feb 23 - Feb 27, 2026

A curated summary of the top protein engineering and structure prediction signals from Feb 23 - Feb 27, 2026.

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.

🧬 Weekly Recap

Feb 23 - Feb 27, 2026

Missed a day? Here are the top research signals and tools from Monday to Friday, summarized in one place.

🏆 Top Signals of the Week

🗓️ Monday, Feb 23

Predicting the active sites of quinolone antibiotics interacting with organisms by deep learning and molecular docking.

🧬 Abstract

Quinolones (QNs) antibiotics have become one of the most commonly used antibacterial drugs for human and animals in the world. In this study, we focused on 19 common quinolone (QN) antibiotics and collected their bioassay activity data from the PubChem website. Subsequently, using deep learning techniques, we constructed 45 biological activity prediction models based on the PubChem BioAssay dataset. The prediction accuracy of all models exceeded 95%, with the exception of the model for CCRIS mutagenicity studies, which achieved an accuracy of 85.22 ± 0.17%. Collectively, these deep learning models can serve as reliable tools for the prediction and evaluation of quinolone antibiotics. The bioassay activity of 19 QNs antibiotics was predicted by developed models to fill in the missing activity data. It was found that QNs antibiotics were generally active against bacterial DNA repair enzymes and neurobehavioral related protein, including hypothetical protein HP1089, recBCD - exodeoxyribonuclease V subunit RecBCD, recombination protein RecB and SLC5A7. Molecular dynamics simulation results showed that all fluoroquinolone complexes with HP1089, recBCD, RecB, and SLC5A7 reached stable conformations after an initial 0-10 ns relaxation, Our research provides a theoretical basis and technical support for elucidating the regulatory mechanisms of organisms in response to environmental exogenous chemicals, the formulation of environmental protection and food safety policies, the risk assessment of novel compounds, and the development of eco-friendly pharmaceuticals.

🗓️ Tuesday, Feb 24

Self-Aware Object Detection via Degradation Manifolds

🧬 Abstract

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.

🗓️ Wednesday, Feb 25

ManCAR: Manifold-Constrained Latent Reasoning with Adaptive Test-Time Computation for Sequential Recommendation

🧬 Abstract

Sequential recommendation increasingly employs latent multi-step reasoning to enhance test-time computation. Despite empirical gains, existing approaches largely drive intermediate reasoning states via target-dominant objectives without imposing explicit feasibility constraints. This results in latent drift, where reasoning trajectories deviate into implausible regions. We argue that effective recommendation reasoning should instead be viewed as navigation on a collaborative manifold rather than free-form latent refinement. To this end, we propose ManCAR (Manifold-Constrained Adaptive Reasoning), a principled framework that grounds reasoning within the topology of a global interaction graph. ManCAR constructs a local intent prior from the collaborative neighborhood of a user’s recent actions, represented as a distribution over the item simplex. During training, the model progressively aligns its latent predictive distribution with this prior, forcing the reasoning trajectory to remain within the valid manifold. At test time, reasoning proceeds adaptively until the predictive distribution stabilizes, avoiding over-refinement. We provide a variational interpretation of ManCAR to theoretically validate its drift-prevention and adaptive test-time stopping mechanisms. Experiments on seven benchmarks demonstrate that ManCAR consistently outperforms state-of-the-art baselines, achieving up to a 46.88% relative improvement w.r.t. NDCG@10. Our code is available at https://github.com/FuCongResearchSquad/ManCAR.

🗓️ Thursday, Feb 26

A Neural Time-Series Learning Method for Accelerating Free-Energy Perturbation and Rare-Event Molecular Dynamics Simulations.

🧬 Abstract

Molecular dynamics (MD) simulations are central to materials and drug discovery yet remain computationally demanding, particularly for free-energy perturbation (FEP) protocols and rare-event sampling. Existing sequence-based accelerators, especially Long Short-Term Memory (LSTM) models, often fail to capture long-range temporal structure and provide sufficient expressive capacity in noisy trajectories. Here, we introduce BiLSTMK-MD, a neural time-series learning method that constructs a causality-preserving surrogate for MD and FEP trajectories to reduce sampling requirements. The approach couples a sliding-window bidirectional LSTM encoder, which preserves long-range correlations, with an attention mechanism to enhance temporally informative frames, while a Kolmogorov-Arnold network output layer applies expressive nonlinear readout to decode the attention-refined representation into the final output. A two-stage, fANOVA-guided Bayesian optimization process searches for the optimal hyperparameter configuration for each system. Across four data sets, BiLSTMK-MD achieves mean absolute errors below 1.5 kcal mol-1 for window-resolved free-energy increments, reconstructs dihedral free-energy basins from 1-10% of trajectories, and maintains performance across systems. In long-trajectory regimes, it affords up to 400-fold acceleration for FEP and >700-fold speedup for rare-conformation sampling over conventional MD/FEP simulation. This neural time-series surrogate therefore provides a route to reducing sampling demands for free-energy estimation and rare-event characterization.

🗓️ Friday, Feb 27

Discovery of potent ALK tyrosine kinase inhibitors for thyroid cancer via machine learning modeling, molecular docking, MD simulations, and DFT study.

🧬 Abstract

The ever-increasing need for effective therapeutic management of thyroid cancer (TC) necessitates the exploration of novel approaches for advanced drug discovery. The current study employed a robust computational pipeline integrating Machine Learning (ML) algorithms, QSAR modeling, molecular docking, molecular dynamics (MD), density functional theory (DFT), and network pharmacology to identify novel Anaplastic Lymphoma Kinase (ALK) tyrosine kinase inhibitors. An initial library of 3546 compounds from the CHEMBL4247 database was systematically filtered to 578. This screening utilized Lipinski’s rule of five, aided by QSAR and detailed PaDEL descriptor analysis. An ensemble ML model, specifically a Voting Classifier (VC) combining XGBoost, LightGBM, and ExtraTrees algorithms, attained high predictive accuracy (ROC-AUC = 0.99), facilitating a strong classification and prioritization of active leads. Molecular docking experiment identified five top hit ligands (60, 63, 124, 130, 204) having docking score ranging from -9.0 to -10.4 kcal/mol and also confirmed their strong binding affinities, which surpassed the native co-crystallized ligand used as a standard. Later on, ADMET studies were executed to explore their physicochemical properties. MD simulation trajectories and MM/PBSA analyses validated the notably conformational stability and favorable binding free energies of these hit complexes. Network pharmacology was incorporated to understand tentative mechanisms of action and potential off-targets, generating a protein-protein interaction (PPI) network. DFT-based frontier molecular orbital (FMO) analysis showed Ligand124 possessed the highest electrophilicity and optimal polarizability, consistent with its marked interaction stability in MD simulations. In addition, the molecular mechanisms of hit compounds against TC were elucidated using a network pharmacology approach, which revealed a compound-target network with crucial hub targets like AKT1 and TP53. Significant correlations with cancer-related pathways, such as PI3K-Akt and MAPK signaling, as well as key involvement in kinase activity, phosphorylation, and membrane signaling complexes, were observed by the enrichment analysis of the main targets. These comprehensive results imply that investigated hit compounds probably modulate the oncogenic signaling networks, especially those controlling cell survival, proliferation, and drug resistance, in order to achieve its anti-TC therapeutic actions. These findings highlight the fundamental ability of integrating ML and computational chemistry to accelerate therapeutic development for TC.

Why it matters: Enhances small-molecule or peptide docking accuracy for targeted drug discovery.

📚 All Papers & Quick Reads

🗓️ Monday, Feb 23

🗓️ Tuesday, Feb 24

🗓️ Wednesday, Feb 25

🗓️ Thursday, Feb 26

🗓️ Friday, Feb 27

🛠️ Tools & Datasets

  • 🛠 Tool: GROMACS - High-performance molecular dynamics engine.
  • 🛠 Tool: OpenMM - GPU-accelerated molecular simulation toolkit.
  • 💾 Dataset: UniRef - Clustered protein sequence sets for fast similarity searches.
  • 💾 Dataset: BFD - Big Fantastic Database for deep learning protein modeling.
  • 🛠 Tool: AlphaFill - Ligand and cofactor transfer into AlphaFold models.
  • 💾 Dataset: MGnify - Metagenomics resource for microbiome sequence data.
  • 🛠 Tool: ReFOLD4 - Sophisticated protein structure refinement tool for improving model quality.
  • 💾 Dataset: PDBbind - Binding affinity data with 3D structures of protein-ligand complexes.
  • 🛠 Tool: FunFOLD5 - Automated system for protein ligand-binding site prediction and function annotation.
  • 💾 Dataset: BioLiP - Verified biologically relevant ligand-protein interactions.
  • 🛠 Tool: MultiFOLD/IntFOLD - High-performance protein structure prediction and quality assessment server.
  • 💾 Dataset: SIFTS - Residue-level mapping between PDB, UniProt, and other resources.

🤖 AI in Research Recap

🏢 Industry & Real-World Applications

💼 Jobs & Opportunities

📅 Events

Enjoyed this digest? Subscribe above to get these dailies in your inbox every morning.

BS HF DK