Selected work

Representative Papers

A Linear-Transformer Hybrid for SNP-Based Genotype-to-Phenotype Prediction in Grapevine

May 07, 2026

Predicting complex phenotypes such as grapevine leaf trichome density from SNP data remains challenging in variable field environments and across years due to limited model robustness. This work proposes LiT-G2P, a novel framework that uniquely integrates linear models—capturing additive genetic effects—with a Transformer architecture to model nonlinear SNP–SNP interactions. Leveraging genome-wide SNP data, attention mechanisms, and genotype-stratified analysis, LiT-G2P achieves single-year and cross-year root mean square errors (RMSE) of 0.469 and 0.454, respectively, corresponding to prediction accuracies of 79.2% and 74.6%, outperforming existing baselines. Moreover, the model’s attention weights enable identification of biologically interpretable candidate functional SNP markers, enhancing both predictive performance and genomic interpretability in perennial crop breeding.

0 citationsRead paper

A two-parameter, minimal-data model to predict dengue cases: the 2022-2023 outbreak in Florida, USA

Nov 26, 2025

Early dengue fever warning in data-sparse regions remains challenging due to limited surveillance infrastructure and scarce covariate data. Method: This paper proposes a minimal two-parameter time-series-only model grounded in a Data-Parsimonious (DP) framework. It integrates incidence curve characterization (ICC analysis), a two-group SEIR dynamical structure, and Bayesian uncertainty quantification to ensure parameter identifiability and robustness—requiring only two estimable parameters, thereby reducing noise sensitivity and computational overhead while yielding well-calibrated prediction intervals. Results: Evaluated on the 2022–2023 Florida dengue outbreak, the model achieves short-term forecasting accuracy comparable to complex covariate-dependent models, yet with over an order-of-magnitude reduction in computational cost. Its core contribution is the first integration of ICC-based curve analysis with a parsimonious SEIR formulation to establish a purely time-series-driven Bayesian dengue forecasting paradigm—delivering a highly practical, real-time monitoring and intervention tool for resource-constrained settings.

0 citationsRead paper

A Data-Parsimonious Model for Long-Term Risk Assessments of West Nile Virus Spillover

Oct 15, 2025

To address the scarcity of entomological and avian surveillance data for West Nile virus (WNV) forecasting, this study proposes a data-light probabilistic framework that relies solely on readily available temperature and mosquito abundance data to enable long-term (multi-month) pre-season risk prediction. Methodologically, we integrate a temperature-driven compartmental transmission model with nonparametric kernel density estimation to jointly estimate the probability density and Poisson intensity surface—enabling robust cross-ecoclimatic extrapolation. Validation across six counties in California, Texas, and Florida demonstrates superior performance in outbreak timing prediction, seasonal intensity quantification, and early warning capability. The framework substantially reduces data requirements, exhibits strong generalizability across diverse ecological settings, and offers direct utility for public health decision-making and resource allocation.

0 citationsRead paper

Ordered Leaf Attachment (OLA) Vectors can Identify Reticulation Events even in Multifurcated Trees

Sep 19, 2025

Identifying reticulate evolutionary events in phylogenetic networks remains computationally challenging, particularly due to the exponential complexity of computing the maximum acyclic agreement forest (MAAF), a key measure for quantifying network discordance. Method: We establish a theoretical connection between ordered leaf attachment (OLA) vectors and the MAAF problem. We introduce a refined OLA distance metric and prove—under an optimal leaf ordering—that it equals the MAAF size exactly; moreover, this distance is computable in linear time. Leveraging this insight, we design a leaf-order-optimized multifurcating tree decomposition algorithm that enables exact MAAF reconstruction. Contribution/Results: Our approach breaks the exponential barrier of traditional MAAF solvers by achieving polynomial-time reticulation event identification. Empirical evaluation demonstrates high efficiency and robustness on microbial datasets incorporating sampling-time information. The framework provides a scalable, theoretically grounded paradigm for large-scale phylogenetic network inference.

0 citationsRead paper

MatchPlant: An Open-Source Pipeline for UAV-Based Single-Plant Detection and Data Extraction

Jun 14, 2025

Accurate single-plant detection in UAV imagery remains challenging due to geometric distortions, scale variation, and labor-intensive annotation. Method: We propose an open-source, modular, GUI-driven pipeline integrating OpenCV-based preprocessing, YOLO/RetinaNet object detection, orthomosaic geometric rectification, IoU-aware geospatial projection, and GDAL-based vectorization. It introduces interactive annotation with cross-temporal detection result reuse and automated plant-height/NDVI extraction. Contribution/Results: Our method achieves seamless, high-accuracy mapping from detection bounding boxes to geographic space (87.5% of projections yield IoU > 0.5) and embeds geospatial analytics. Evaluated on early-stage maize, it attains AP = 89.6% (85.9% on test set), with phenotypic parameters strongly correlated (r = 0.87–0.97) against ground truth and covering 89.8% of manual annotations—substantially reducing annotation effort and cost.

0 citationsRead paper

Recent publications

Latest Papers

A Linear-Transformer Hybrid for SNP-Based Genotype-to-Phenotype Prediction in Grapevine

May 07, 2026

Predicting complex phenotypes such as grapevine leaf trichome density from SNP data remains challenging in variable field environments and across years due to limited model robustness. This work proposes LiT-G2P, a novel framework that uniquely integrates linear models—capturing additive genetic effects—with a Transformer architecture to model nonlinear SNP–SNP interactions. Leveraging genome-wide SNP data, attention mechanisms, and genotype-stratified analysis, LiT-G2P achieves single-year and cross-year root mean square errors (RMSE) of 0.469 and 0.454, respectively, corresponding to prediction accuracies of 79.2% and 74.6%, outperforming existing baselines. Moreover, the model’s attention weights enable identification of biologically interpretable candidate functional SNP markers, enhancing both predictive performance and genomic interpretability in perennial crop breeding.

0 citationsRead paper

A two-parameter, minimal-data model to predict dengue cases: the 2022-2023 outbreak in Florida, USA

Nov 26, 2025

Early dengue fever warning in data-sparse regions remains challenging due to limited surveillance infrastructure and scarce covariate data. Method: This paper proposes a minimal two-parameter time-series-only model grounded in a Data-Parsimonious (DP) framework. It integrates incidence curve characterization (ICC analysis), a two-group SEIR dynamical structure, and Bayesian uncertainty quantification to ensure parameter identifiability and robustness—requiring only two estimable parameters, thereby reducing noise sensitivity and computational overhead while yielding well-calibrated prediction intervals. Results: Evaluated on the 2022–2023 Florida dengue outbreak, the model achieves short-term forecasting accuracy comparable to complex covariate-dependent models, yet with over an order-of-magnitude reduction in computational cost. Its core contribution is the first integration of ICC-based curve analysis with a parsimonious SEIR formulation to establish a purely time-series-driven Bayesian dengue forecasting paradigm—delivering a highly practical, real-time monitoring and intervention tool for resource-constrained settings.

0 citationsRead paper

A Data-Parsimonious Model for Long-Term Risk Assessments of West Nile Virus Spillover

Oct 15, 2025

To address the scarcity of entomological and avian surveillance data for West Nile virus (WNV) forecasting, this study proposes a data-light probabilistic framework that relies solely on readily available temperature and mosquito abundance data to enable long-term (multi-month) pre-season risk prediction. Methodologically, we integrate a temperature-driven compartmental transmission model with nonparametric kernel density estimation to jointly estimate the probability density and Poisson intensity surface—enabling robust cross-ecoclimatic extrapolation. Validation across six counties in California, Texas, and Florida demonstrates superior performance in outbreak timing prediction, seasonal intensity quantification, and early warning capability. The framework substantially reduces data requirements, exhibits strong generalizability across diverse ecological settings, and offers direct utility for public health decision-making and resource allocation.

0 citationsRead paper

Ordered Leaf Attachment (OLA) Vectors can Identify Reticulation Events even in Multifurcated Trees

Sep 19, 2025

Identifying reticulate evolutionary events in phylogenetic networks remains computationally challenging, particularly due to the exponential complexity of computing the maximum acyclic agreement forest (MAAF), a key measure for quantifying network discordance. Method: We establish a theoretical connection between ordered leaf attachment (OLA) vectors and the MAAF problem. We introduce a refined OLA distance metric and prove—under an optimal leaf ordering—that it equals the MAAF size exactly; moreover, this distance is computable in linear time. Leveraging this insight, we design a leaf-order-optimized multifurcating tree decomposition algorithm that enables exact MAAF reconstruction. Contribution/Results: Our approach breaks the exponential barrier of traditional MAAF solvers by achieving polynomial-time reticulation event identification. Empirical evaluation demonstrates high efficiency and robustness on microbial datasets incorporating sampling-time information. The framework provides a scalable, theoretically grounded paradigm for large-scale phylogenetic network inference.

0 citationsRead paper

MatchPlant: An Open-Source Pipeline for UAV-Based Single-Plant Detection and Data Extraction

Jun 14, 2025

Accurate single-plant detection in UAV imagery remains challenging due to geometric distortions, scale variation, and labor-intensive annotation. Method: We propose an open-source, modular, GUI-driven pipeline integrating OpenCV-based preprocessing, YOLO/RetinaNet object detection, orthomosaic geometric rectification, IoU-aware geospatial projection, and GDAL-based vectorization. It introduces interactive annotation with cross-temporal detection result reuse and automated plant-height/NDVI extraction. Contribution/Results: Our method achieves seamless, high-accuracy mapping from detection bounding boxes to geographic space (87.5% of projections yield IoU > 0.5) and embeds geospatial analytics. Evaluated on early-stage maize, it attains AP = 89.6% (85.9% on test set), with phenotypic parameters strongly correlated (r = 0.87–0.97) against ground truth and covering 89.8% of manual annotations—substantially reducing annotation effort and cost.

0 citationsRead paper