Ecir Uğur Küçüksille
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ecir Uğur Küçüksille has a total of 116 co-authors. Here are the top co-authors by number of shared publications:
- Kevser Kübra Kırboğa
Faculty of Engineering, Department of Bioengineering Bilecik Şeyh Edebali University Bilecik Türkiye
7 shared publications - Tülay Turan
Department of Computer Engineering Faculty of Engineering and Architecture Burdur Mehmet Akif Ersoy University Burdur Turkiye
5 shared publications - Nazan Kemaloğlu Alagöz
Isparta Uygulamalı Bilimler Üniversitesi
4 shared publications - Güncel Sarıman
Mugla Sitki Kocman Universitesi
3 shared publications Mehmet Ali Yalçınkaya
3 shared publications- Merve Varol Arısoy
Bucak Faculty of Computer and Informatics, Information Systems Engineering Department, Burdur Mehmet Akif Ersoy University, Burdur, Turkey
3 shared publications
Latest Publications
- Open Access
- 25 October 2026
- Published by Anatolian Current Medical Journal in Anatolian Current Medical Journal
- Vol. 8 (6) , 1262-1268
Registered period prevalence of celiac disease in Turkiye, 2019–2024: regional and age-specific distribution and the diagnostic gap
Aims: To estimate the registered period prevalence of celiac disease in Turkiye during 2019-2024, describe regional and agespecific patterns, and compare registry-based estimates with screening-based prevalence.Methods: This nationwide, registry-based descriptive study used a full-census design. Individuals with an ICD-10 K90.0 diagnosis in the Ministry of Health electronic health registry were identified and deduplicated using encrypted national identifiers. Turkish Statistical Institute population data were used as denominators. Because diagnoses accumulated over six years, estimates represent period prevalence. Prevalence was calculated for all records and separately for confirmed (main) diagnoses; no inferential testing was performed.Results: Among 98.028 patients, 41.125 (41.95%) were children and 56.903 (58.05%) adults; 63.74% of pediatric and 60.78% of adult patients had confirmed diagnoses. Registered period prevalence per 100.000 population was 113.9 overall and 70.6 for confirmed records. Corresponding estimates were 181.4 and 115.6 in children and 89.7 and 54.5 in adults. Registered case rates varied more than twofold across regions of care, with the highest rate in Central Anatolia and the lowest in the Black Sea region. Laboratory panel coverage ranged from 6.95% to 93.31%, and abnormality rates varied substantially according to the denominator used.Conclusion: Registered prevalence was higher in children than adults and was roughly one-tenth of screening-based estimates (about 1%); differences in period, coverage and case definition preclude direct quantification of the gap. Variation in regional registered case rates and the substantial reduction in cases when only confirmed diagnoses were included may reflect incomplete case detection and inconsistent diagnostic coding.
- 25 October 2026
- Published by Anatolian Current Medical Journal in Anatolian Current Medical Journal
- Vol. 8 (6) , 1262-1268
Open AccessRegistered period prevalence of celiac disease in Turkiye, 2019–2024: regional and age-specific distribution and the diagnostic gap
Aims: To estimate the registered period prevalence of celiac disease in Turkiye during 2019-2024, describe regional and agespecific patterns, and compare registry-based estimates with screening-based prevalence.Methods: This nationwide, registry-based descriptive study used a full-census design. Individuals with an ICD-10 K90.0 diagnosis in the Ministry of Health electronic health registry were identified and deduplicated using encrypted national identifiers. Turkish Statistical Institute population data were used as denominators. Because diagnoses accumulated over six years, estimates represent period prevalence. Prevalence was calculated for all records and separately for confirmed (main) diagnoses; no inferential testing was performed.Results: Among 98.028 patients, 41.125 (41.95%) were children and 56.903 (58.05%) adults; 63.74% of pediatric and 60.78% of adult patients had confirmed diagnoses. Registered period prevalence per 100.000 population was 113.9 overall and 70.6 for confirmed records. Corresponding estimates were 181.4 and 115.6 in children and 89.7 and 54.5 in adults. Registered case rates varied more than twofold across regions of care, with the highest rate in Central Anatolia and the lowest in the Black Sea region. Laboratory panel coverage ranged from 6.95% to 93.31%, and abnormality rates varied substantially according to the denominator used.Conclusion: Registered prevalence was higher in children than adults and was roughly one-tenth of screening-based estimates (about 1%); differences in period, coverage and case definition preclude direct quantification of the gap. Variation in regional registered case rates and the substantial reduction in cases when only confirmed diagnoses were included may reflect incomplete case detection and inconsistent diagnostic coding.
- Open Access
- 1 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (19) , 4506
Quantum Autoencoders for Network Anomaly Detection: A Comparative Benchmark with Classical Methods
The practical value of quantum machine learning for network anomaly detection remains unclear under controlled and comparable benchmarking conditions. This study evaluates a trash-qubit-based Quantum Autoencoder (QAE) against seven classical anomaly-detection methods on UNSW-NB15 using shared data partitions, preprocessing steps, model selection procedures, and threshold calibration protocols. The experimental design included five independent seeds, three training set sizes, two feature budgets, and 12 QAE architectures, together with low-FPR analysis, finite-shot and synthetic-noise evaluation, less-constrained classical reference settings, and a cross-dataset assessment on TON_IoT. The Angle-RY encoding with a Real-Amplitudes Ring ansatz at depth L = 2 achieved the highest mean validation-selection AUPRC. On UNSW-NB15, the QAE was competitive in AUPRC and AUROC but remained behind the strongest classical baselines in F1-score, and its attack recall declined sharply under low-FPR constraints. The TON_IoT results further showed that relative model performance depended on the dataset. Under finite-shot and synthetic-noise conditions, anomaly-score rankings were largely preserved, whereas threshold-based decisions were more sensitive. Overall, the QAE showed competitive continuous-score discrimination on UNSW-NB15, but achieving reliable detection at low false-alarm rates remains a key operational challenge.
- 1 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (19) , 4506
Open AccessQuantum Autoencoders for Network Anomaly Detection: A Comparative Benchmark with Classical Methods
The practical value of quantum machine learning for network anomaly detection remains unclear under controlled and comparable benchmarking conditions. This study evaluates a trash-qubit-based Quantum Autoencoder (QAE) against seven classical anomaly-detection methods on UNSW-NB15 using shared data partitions, preprocessing steps, model selection procedures, and threshold calibration protocols. The experimental design included five independent seeds, three training set sizes, two feature budgets, and 12 QAE architectures, together with low-FPR analysis, finite-shot and synthetic-noise evaluation, less-constrained classical reference settings, and a cross-dataset assessment on TON_IoT. The Angle-RY encoding with a Real-Amplitudes Ring ansatz at depth L = 2 achieved the highest mean validation-selection AUPRC. On UNSW-NB15, the QAE was competitive in AUPRC and AUROC but remained behind the strongest classical baselines in F1-score, and its attack recall declined sharply under low-FPR constraints. The TON_IoT results further showed that relative model performance depended on the dataset. Under finite-shot and synthetic-noise conditions, anomaly-score rankings were largely preserved, whereas threshold-based decisions were more sensitive. Overall, the QAE showed competitive continuous-score discrimination on UNSW-NB15, but achieving reliable detection at low false-alarm rates remains a key operational challenge.
- Open Access
- 11 September 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
From DevOps to XOps: an agent-driven reference architecture for autonomous enterprise operations
Enterprise adoption of machine learning has fragmented operations into specialised disciplines—DataOps, MLOps, AIOps—creating silos that impede unified governance. We propose XOps, a five-layer reference architecture integrating PlatformOps, DataOps, MLOps and AIOps beneath an Agentic Orchestration layer with Policy-as-Code governance, together with a continuous-time Markov chain model quantifying the availability effect of agent-driven remediation. Two case studies evaluate the architecture under controlled, synthetic conditions. For a self-healing payment gateway, 250 live executions of the reasoning graph against a hosted language model and a live policy engine yield 85.6% plan-level action consistency and 99.6% fault classification accuracy, with 12.4% of plans rejected by the policy gate and escalated to a human and no policy-violating action authorised for execution; the pipeline’s 3.3-minute recovery time is a simulated latency budget rather than a cluster measurement. For a predictive-maintenance application on NASA C-MAPSS data, autonomous drift detection and retraining sustain $$R^2 = 0.74$$ against 0.29 for an equivalent static model, measured on engines reserved entirely from retraining. An indicative cost analysis suggests an approximately 70% reduction in expected monthly operational cost. Within this scope the results support the feasibility of agent-driven operations rather than establishing production-scale performance.- 11 September 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessFrom DevOps to XOps: an agent-driven reference architecture for autonomous enterprise operations
Enterprise adoption of machine learning has fragmented operations into specialised disciplines—DataOps, MLOps, AIOps—creating silos that impede unified governance. We propose XOps, a five-layer reference architecture integrating PlatformOps, DataOps, MLOps and AIOps beneath an Agentic Orchestration layer with Policy-as-Code governance, together with a continuous-time Markov chain model quantifying the availability effect of agent-driven remediation. Two case studies evaluate the architecture under controlled, synthetic conditions. For a self-healing payment gateway, 250 live executions of the reasoning graph against a hosted language model and a live policy engine yield 85.6% plan-level action consistency and 99.6% fault classification accuracy, with 12.4% of plans rejected by the policy gate and escalated to a human and no policy-violating action authorised for execution; the pipeline’s 3.3-minute recovery time is a simulated latency budget rather than a cluster measurement. For a predictive-maintenance application on NASA C-MAPSS data, autonomous drift detection and retraining sustain $$R^2 = 0.74$$ against 0.29 for an equivalent static model, measured on engines reserved entirely from retraining. An indicative cost analysis suggests an approximately 70% reduction in expected monthly operational cost. Within this scope the results support the feasibility of agent-driven operations rather than establishing production-scale performance. - 1 September 2026
- Published by Gazi University Journal of Science in Gazi University Journal of Science
- Vol. 39 (3) , 1492-1521
Time Series Forecasting of Greenhouse Gas Emissions Using Deep Learning
Greenhouse gas (GHG) emissions remain a primary driver of global climate change, and accurate forecasting is critical for evaluating climate policies and supporting sustainable development goals. This study conducts a time series analysis in Python using the EDGAR dataset for 1970–2023, explicitly adopting a sectoral scope that covers Agriculture, Buildings, Fuel Exploitation, Industrial Combustion, Power Industry, Processes, Transport, and Waste. Long Short-Term Memory (LSTM) models were developed and evaluated via rolling/expanding-window backtesting, while Monte Carlo Dropout (MCD) was applied to quantify predictive uncertainty. Out-of-distribution (OOD) tests were further used to assess generalization under distributional shifts, and early stopping with learning-rate scheduling was employed to mitigate overfitting. While the LSTM captures the dominant long-term upward trend, its out-of-sample performance is constrained under regime changes and shocks (Test RMSE = 1228.66; MAE = 991.35; R2=−1.0020) and it underperforms a naïve benchmark. Nested rolling-origin results also indicate rapidly increasing errors at longer horizons. In contrast, a Transformer specification improves performance over the last five years (RMSE = 777.21; MAE = 663.39; R2=0.1989). Sectoral findings suggest that short-term declines around 2020 are concentrated in transport and energy-related sectors, while agriculture and industrial/process-related emissions remain relatively stable.
- 1 September 2026
- Published by Gazi University Journal of Science in Gazi University Journal of Science
- Vol. 39 (3) , 1492-1521
Time Series Forecasting of Greenhouse Gas Emissions Using Deep Learning
Greenhouse gas (GHG) emissions remain a primary driver of global climate change, and accurate forecasting is critical for evaluating climate policies and supporting sustainable development goals. This study conducts a time series analysis in Python using the EDGAR dataset for 1970–2023, explicitly adopting a sectoral scope that covers Agriculture, Buildings, Fuel Exploitation, Industrial Combustion, Power Industry, Processes, Transport, and Waste. Long Short-Term Memory (LSTM) models were developed and evaluated via rolling/expanding-window backtesting, while Monte Carlo Dropout (MCD) was applied to quantify predictive uncertainty. Out-of-distribution (OOD) tests were further used to assess generalization under distributional shifts, and early stopping with learning-rate scheduling was employed to mitigate overfitting. While the LSTM captures the dominant long-term upward trend, its out-of-sample performance is constrained under regime changes and shocks (Test RMSE = 1228.66; MAE = 991.35; R2=−1.0020) and it underperforms a naïve benchmark. Nested rolling-origin results also indicate rapidly increasing errors at longer horizons. In contrast, a Transformer specification improves performance over the last five years (RMSE = 777.21; MAE = 663.39; R2=0.1989). Sectoral findings suggest that short-term declines around 2020 are concentrated in transport and energy-related sectors, while agriculture and industrial/process-related emissions remain relatively stable.
- 8 August 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (12) , 609
SeqGPT-RL: a reinforcement-aligned hybrid intelligent system for Turkish technical-support dialogs
- 8 August 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (12) , 609
SeqGPT-RL: a reinforcement-aligned hybrid intelligent system for Turkish technical-support dialogs
- 25 June 2026
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- No. Advanced O,p. 958-979
Signal Processing and Dimensionality Reduction Algorithms for High-Dimensional Time-Series Data Applied to Biomolecular Simulations
High-dimensional time-series datasets with stochastic noise present fundamental challenges in signal processing and data analytics across diverse computational domains. This study develops and validates a systematic framework that integrates digital signal processing techniques with dimensionality reduction algorithms to extract meaningful trends from noisy, high-dimensional temporal data. We implement and compare five filtering approaches—Savitzky-Golay polynomial regression, Moving Average, Gaussian, Butterworth, and Wavelet denoising—combined with Principal Component Analysis (PCA) on GPU-accelerated MD simulation infrastructure (NVIDIA GPU based on the Blackwell architecture, CUDA 12.9), benchmarking performance on two contrasting biomolecular systems: Frataxin (119 residues, rigid) and Carbonic Anhydrase IX (CAIX) (257 residues, dynamic). A comprehensive five-filter comparison across both systems reveals consistent performance rankings: Wavelet achieves the highest signal-to-noise ratio (SNR) for both Frataxin (23.47 dB) and CAIX (13.80 dB), while Savitzky-Golay provides the optimal balance between noise reduction and low-frequency preservation (>99%). Critically, Savitzky-Golay’s improvement over the Moving Average is substantially greater for dynamic CAIX (16.3%) than for rigid Frataxin (3.1%), demonstrating enhanced performance precisely where distinguishing conformational transitions from thermal noise is most challenging. All filters preserve low-frequency conformational dynamics while reducing high-frequency noise, with the extracted noise components validated by Gaussian distribution analysis (σ ≈ 0.008 nm) across both systems and all filtering methods. PCA-based dimensionality reduction achieves 11.9:1 compression (357 dimensions → 30 principal components) for Frataxin while retaining 80% of conformational variance, with CAIX requiring only 11 components for equivalent variance capture due to dominant collective motions. The complete analysis pipeline processes 50,001 frames in less than 0.2 seconds, representing negligible overhead (<0.001%) relative to simulation time. Cross-system validation with consistent filter rankings confirms methodology generalizability across proteins spanning diverse dynamical regimes. All analysis pipelines are implemented in Python 3.13 with open-source libraries (NumPy, SciPy, Matplotlib) to ensure reproducibility and extensibility.
- 25 June 2026
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- No. Advanced O,p. 958-979
Signal Processing and Dimensionality Reduction Algorithms for High-Dimensional Time-Series Data Applied to Biomolecular Simulations
High-dimensional time-series datasets with stochastic noise present fundamental challenges in signal processing and data analytics across diverse computational domains. This study develops and validates a systematic framework that integrates digital signal processing techniques with dimensionality reduction algorithms to extract meaningful trends from noisy, high-dimensional temporal data. We implement and compare five filtering approaches—Savitzky-Golay polynomial regression, Moving Average, Gaussian, Butterworth, and Wavelet denoising—combined with Principal Component Analysis (PCA) on GPU-accelerated MD simulation infrastructure (NVIDIA GPU based on the Blackwell architecture, CUDA 12.9), benchmarking performance on two contrasting biomolecular systems: Frataxin (119 residues, rigid) and Carbonic Anhydrase IX (CAIX) (257 residues, dynamic). A comprehensive five-filter comparison across both systems reveals consistent performance rankings: Wavelet achieves the highest signal-to-noise ratio (SNR) for both Frataxin (23.47 dB) and CAIX (13.80 dB), while Savitzky-Golay provides the optimal balance between noise reduction and low-frequency preservation (>99%). Critically, Savitzky-Golay’s improvement over the Moving Average is substantially greater for dynamic CAIX (16.3%) than for rigid Frataxin (3.1%), demonstrating enhanced performance precisely where distinguishing conformational transitions from thermal noise is most challenging. All filters preserve low-frequency conformational dynamics while reducing high-frequency noise, with the extracted noise components validated by Gaussian distribution analysis (σ ≈ 0.008 nm) across both systems and all filtering methods. PCA-based dimensionality reduction achieves 11.9:1 compression (357 dimensions → 30 principal components) for Frataxin while retaining 80% of conformational variance, with CAIX requiring only 11 components for equivalent variance capture due to dominant collective motions. The complete analysis pipeline processes 50,001 frames in less than 0.2 seconds, representing negligible overhead (<0.001%) relative to simulation time. Cross-system validation with consistent filter rankings confirms methodology generalizability across proteins spanning diverse dynamical regimes. All analysis pipelines are implemented in Python 3.13 with open-source libraries (NumPy, SciPy, Matplotlib) to ensure reproducibility and extensibility.
- Open Access
- 2 June 2026
- Published by International Journal of Engineering and Innovative Research in International Journal of Engineering and Innovative Research
- Vol. 8 (1) , 1-10
PREDICTION OF THE NUMBER OF FATAL OCCUPATIONAL ACCIDENTS IN THE TEXTILE INDUSTRY
The primary objective of this study is to identify the number of occupational accidents occurring in the Turkish textile sector, specifically within the domains of "textile manufacturing" and "apparel production." A comprehensive review of the literature reveals the absence of studies that analyze occupational health and safety risks in the Turkish textile industry using data from 2017 to 2023. To address this gap, a data-driven fuzzy logic modeling approach was applied to accident data obtained from the Social Security Institution of Turkey (SGK). The ultimate aim of the study is to forecast the number of fatal workplace accidents in the textile sector for the year 2024 data that has yet to be officially released based on historical data from 2017 to 2023. Fuzzy logic methods were employed as the forecasting technique. The validity of these predictions will be assessed once the 2024 accident data becomes available. Additionally, the article briefly reviews the application of fuzzy logic over the past eight years in estimating key occupational safety indicators such as the number of accidents, number of occupational diseases, and number of deaths.
- 2 June 2026
- Published by International Journal of Engineering and Innovative Research in International Journal of Engineering and Innovative Research
- Vol. 8 (1) , 1-10
Open AccessPREDICTION OF THE NUMBER OF FATAL OCCUPATIONAL ACCIDENTS IN THE TEXTILE INDUSTRY
The primary objective of this study is to identify the number of occupational accidents occurring in the Turkish textile sector, specifically within the domains of "textile manufacturing" and "apparel production." A comprehensive review of the literature reveals the absence of studies that analyze occupational health and safety risks in the Turkish textile industry using data from 2017 to 2023. To address this gap, a data-driven fuzzy logic modeling approach was applied to accident data obtained from the Social Security Institution of Turkey (SGK). The ultimate aim of the study is to forecast the number of fatal workplace accidents in the textile sector for the year 2024 data that has yet to be officially released based on historical data from 2017 to 2023. Fuzzy logic methods were employed as the forecasting technique. The validity of these predictions will be assessed once the 2024 accident data becomes available. Additionally, the article briefly reviews the application of fuzzy logic over the past eight years in estimating key occupational safety indicators such as the number of accidents, number of occupational diseases, and number of deaths.
- Open Access
Structure-Based Virtual Screening of Plant-Derived Flavonoids as Putative GLUT9 Binders with Antioxidant Properties
Hyperuricemia affects approximately 20% of the global adult population and serves as the primary etiological factor for gout. Glucose transporter 9 (GLUT9) plays a critical role in renal urate reabsorption, representing a promising therapeutic target for hyperuricemia treatment. This study employed an integrated computational and experimental approach to identify novel flavonoid-based putative GLUT9 binders, combining molecular docking, molecular dynamics (MD) simulations, ADMET prediction, antioxidant evaluation, and density functional theory (DFT) calculations. Eight structurally diverse flavonoids were docked against the human GLUT9 cryo-EM structure, and antioxidant activities were assessed using DPPH, ABTS, and FRAP assays. All tested flavonoids exhibited favorable binding affinities ranging from −7.67 to −9.10 kcal/mol. Epigallocatechin gallate (EGCG) demonstrated the highest binding affinity (−9.10 kcal/mol) with an extensive hydrogen bonding network, while chrysin exhibited the second-highest affinity (−8.35 kcal/mol) with favorable drug-like properties. MD simulations over 100 ns confirmed the structural stability of the complexes. EGCG displayed exceptional antioxidant capacity (DPPH IC50 = 3.28 μM) superior to ascorbic acid, whereas chrysin showed lower radical scavenging activity despite favorable GLUT9 binding. DFT calculations revealed that higher HOMO energies correlated with enhanced antioxidant activity. These findings suggest that EGCG and chrysin exhibit favorable GLUT9 binding profiles, warranting further functional and pharmacokinetic optimization.
Open AccessStructure-Based Virtual Screening of Plant-Derived Flavonoids as Putative GLUT9 Binders with Antioxidant Properties
Hyperuricemia affects approximately 20% of the global adult population and serves as the primary etiological factor for gout. Glucose transporter 9 (GLUT9) plays a critical role in renal urate reabsorption, representing a promising therapeutic target for hyperuricemia treatment. This study employed an integrated computational and experimental approach to identify novel flavonoid-based putative GLUT9 binders, combining molecular docking, molecular dynamics (MD) simulations, ADMET prediction, antioxidant evaluation, and density functional theory (DFT) calculations. Eight structurally diverse flavonoids were docked against the human GLUT9 cryo-EM structure, and antioxidant activities were assessed using DPPH, ABTS, and FRAP assays. All tested flavonoids exhibited favorable binding affinities ranging from −7.67 to −9.10 kcal/mol. Epigallocatechin gallate (EGCG) demonstrated the highest binding affinity (−9.10 kcal/mol) with an extensive hydrogen bonding network, while chrysin exhibited the second-highest affinity (−8.35 kcal/mol) with favorable drug-like properties. MD simulations over 100 ns confirmed the structural stability of the complexes. EGCG displayed exceptional antioxidant capacity (DPPH IC50 = 3.28 μM) superior to ascorbic acid, whereas chrysin showed lower radical scavenging activity despite favorable GLUT9 binding. DFT calculations revealed that higher HOMO energies correlated with enhanced antioxidant activity. These findings suggest that EGCG and chrysin exhibit favorable GLUT9 binding profiles, warranting further functional and pharmacokinetic optimization.
- 21 January 2026
- Published by Wiley in Proteins: Structure, Function, and Bioinformatics
- Vol. 94 (6) , 1225-1244
Integration of Evolutionary Analysis With RFdiffusion for De Novo Design of Aggregation‐Resistant Frataxin
Friedreich's ataxia (FRDA) is a neurodegenerative disorder caused by frataxin (FXN) deficiency, where protein replacement therapy is hampered by the inherent instability and aggregation propensity of wild‐type (WT) FXN. The structural flexibility of Loop‐1 (residues 115–123), a critical region within the acidic ridge, represents a key determinant of protein stability. This study introduces a computational pipeline integrating evolutionary conservation analysis (ConSurf) with diffusion‐based de novo design (RFdiffusion) to redesign both the backbone and sequence of Loop‐1. Through systematic filtration of 1000 ProteinMPNN‐generated variants using aggregation propensity screening (AGGRESCAN) and 450 ns of molecular dynamics (MD) simulations, four lead candidates were identified. Design_188 (EERVGGREI) demonstrated optimal performance with 2.3‐fold improvement in aggregation resistance (Na4vSS: −53.8 vs. −23.5 for WT), superior structural stability (RMSD: 0.486 nm), reduced conformational diversity (62.3% dominant cluster occupancy), and 93% retention of ISCU binding capacity (ΔΔG: +6.4 kcal/mol). Experimental validation through 15 N NMR relaxation analysis confirmed computational predictions, with Design_188 exhibiting uniform backbone rigidification (S 2 = 0.81–0.95) and strong MD‐NMR correlation (Pearson r = 0.675, p = 0.003). SEC‐MALS analysis demonstrated near‐complete monomeric behavior (> 98% monomer content) compared to WT's heterogeneous oligomerization (68% monomer, 32% oligomers), directly confirming the predicted anti‐aggregation properties. K‐means clustering analysis revealed an inverse relationship between conformational heterogeneity and stability, while correlation analysis identified a fundamental trade‐off between aggregation resistance and structural stability ( r = −0.82, p < 0.01). This work establishes a generalizable framework for therapeutic protein engineering where backbone redesign enables conformational ensemble modulation beyond the limitations of sequence optimization alone.
- 21 January 2026
- Published by Wiley in Proteins: Structure, Function, and Bioinformatics
- Vol. 94 (6) , 1225-1244
Integration of Evolutionary Analysis With RFdiffusion for De Novo Design of Aggregation‐Resistant Frataxin
Friedreich's ataxia (FRDA) is a neurodegenerative disorder caused by frataxin (FXN) deficiency, where protein replacement therapy is hampered by the inherent instability and aggregation propensity of wild‐type (WT) FXN. The structural flexibility of Loop‐1 (residues 115–123), a critical region within the acidic ridge, represents a key determinant of protein stability. This study introduces a computational pipeline integrating evolutionary conservation analysis (ConSurf) with diffusion‐based de novo design (RFdiffusion) to redesign both the backbone and sequence of Loop‐1. Through systematic filtration of 1000 ProteinMPNN‐generated variants using aggregation propensity screening (AGGRESCAN) and 450 ns of molecular dynamics (MD) simulations, four lead candidates were identified. Design_188 (EERVGGREI) demonstrated optimal performance with 2.3‐fold improvement in aggregation resistance (Na4vSS: −53.8 vs. −23.5 for WT), superior structural stability (RMSD: 0.486 nm), reduced conformational diversity (62.3% dominant cluster occupancy), and 93% retention of ISCU binding capacity (ΔΔG: +6.4 kcal/mol). Experimental validation through 15 N NMR relaxation analysis confirmed computational predictions, with Design_188 exhibiting uniform backbone rigidification (S 2 = 0.81–0.95) and strong MD‐NMR correlation (Pearson r = 0.675, p = 0.003). SEC‐MALS analysis demonstrated near‐complete monomeric behavior (> 98% monomer content) compared to WT's heterogeneous oligomerization (68% monomer, 32% oligomers), directly confirming the predicted anti‐aggregation properties. K‐means clustering analysis revealed an inverse relationship between conformational heterogeneity and stability, while correlation analysis identified a fundamental trade‐off between aggregation resistance and structural stability ( r = −0.82, p < 0.01). This work establishes a generalizable framework for therapeutic protein engineering where backbone redesign enables conformational ensemble modulation beyond the limitations of sequence optimization alone.