publications

2026

1. arXiv

   BVSIMC: Bayesian Variable Selection-Guided Inductive Matrix Completion for Improved and Interpretable Drug Discovery

   Sijian Fan, Liyan Xiong, Dayuan Wang, Guoshuai Cai, and 1 more author

   2026

   Abs Bib PDF Code

   Recent advances in drug discovery have demonstrated that incorporating side information (e.g., chemical properties about drugs and genomic information about diseases) often greatly improves prediction performance. However, these side features can vary widely in relevance and are often noisy and high-dimensional. We propose Bayesian Variable Selection-Guided Inductive Matrix Completion (BVSIMC), a new Bayesian model that enables variable selection from side features in drug discovery. By learning sparse latent embeddings, BVSIMC improves both predictive accuracy and interpretability. We validate our method through simulation studies and two drug discovery applications: 1) prediction of drug resistance in Mycobacterium tuberculosis, and 2) prediction of new drug-disease associations in computational drug repositioning. On both synthetic and real data, BVSIMC outperforms several other state-of-the-art methods in terms of prediction. In our two real examples, BVSIMC further reveals the most clinically meaningful side features.

   @misc{fan2026bvsimcbayesianvariableselectionguided,
     title = {BVSIMC: Bayesian Variable Selection-Guided Inductive Matrix Completion for Improved and Interpretable Drug Discovery},
     author = {Fan, Sijian and Xiong, Liyan and Wang, Dayuan and Cai, Guoshuai and Bai, Ray},
     year = {2026},
     eprint = {2603.18957},
     archiveprefix = {arXiv},
     primaryclass = {cs.LG},
   }

2. arXiv

   BiSSLB: Binary Spike-and-Slab Lasso Biclustering

   Sijian Fan and Ray Bai

   2026

   Abs Bib PDF Code

   Binary biclustering is a crucial analytical technique for identifying local patterns in binary data matrices, with applications spanning genomics, text mining, and market analysis. In this study, we propose a novel statistical methodology for binary biclustering called Binary Spike-and-Slab Lasso Biclustering (BiSSLB), which enhances both accuracy and interpretability. Our approach is based on a logistic matrix factorization model with spike-and-slab lasso priors that enable adaptive shrinkage of latent spaces to exact zeros. To automatically determine the number of biclusters, we incorporate Indian Buffet Process (IBP) priors, which induce column-wise sparsity in the latent space. Furthermore, we employ a highly efficient coordinate descent method with proximal steps, allowing for scalable estimation in large-scale real-world applications.
   
   To assess the effectiveness of our method, we conduct extensive comparisons against established biclustering techniques, including Bimax, BiBit, iBBiG, and GBC. Performance is evaluated using both simulated datasets and real gene expression datasets, with key metrics including clustering error (CE), consensus score (CS), recovery, relevance, sensitivity, specificity, Matthews correlation coefficient (MCC), and the number of biclusters (K). Experimental results demonstrate that our proposed method accurately estimates the true number of biclusters, regardless of overlapping regions or background noise. Additionally, our method consistently outperforms state-of-the-art approaches in binary biclustering, achieving higher CE, CS, and MCC—three comprehensive metrics for overall comparison.
   
   The proposed methodology advances binary biclustering by offering improved bicluster determination, enhanced robustness to noisy data, and greater interpretability in binary gene expression analysis. Future research directions include integrating side information into the model and extending it to multi-class data, further enhancing its applicability for differential gene expression analysis, as well as broader domains such as recommender systems, drug discovery, and bibliometric studies.

   @misc{fan2026bisslbbinaryspikeandslablasso,
     title = {BiSSLB: Binary Spike-and-Slab Lasso Biclustering},
     author = {Fan, Sijian and Bai, Ray},
     year = {2026},
     eprint = {2603.18378},
     archiveprefix = {arXiv},
     primaryclass = {stat.ME},
   }

2024

1. Caries Res.

   Predictors of Developmental Defects of Enamel in Primary Maxillary Central Incisors Using Bayesian Model Selection

   Susan G Reed, Sijian Fan, Carol L Wagner, and Andrew B Lawson

   Caries research, 2024

   Abs Bib

   Introduction: Localized non-inheritable developmental defects of tooth enamel (DDE) are classified as enamel hypoplasia (EH), opacity (OP), and post-eruptive breakdown (PEB) using the enamel defects index. To better understand the etiology of DDE, we assessed the linkages amongst exposome variables for these defects during the specific time duration for enamel mineralization of the human primary maxillary central incisor enamel crowns. In general, these two teeth develop between 13 and 14 weeks in utero and 3–4 weeks postpartum of a full-term delivery, followed by tooth eruption at about 1 year of age.
   
   Methods: We utilized existing datasets for mother–child dyads that encompassed 12 weeks’ gestation through birth and early infancy, and child DDE outcomes from digital images of the erupted primary maxillary central incisor teeth. We applied a Bayesian modeling paradigm to assess the important predictors of EH, OP, and PEB.
   
   Results: The results of Gibbs variable selection showed a key set of predictors: mother’s prepregnancy body mass index (BMI); maternal serum concentrations of calcium and phosphorus at gestational week 28; child’s gestational age; and both mother’s and child’s functional vitamin D deficiency (FVDD). In this sample of healthy mothers and children, significant predictors for OP included the child having a gestational period greater than 36 weeks and FVDD at birth, and for PEB included a mother’s prepregnancy BMI less than 21.5 and higher serum phosphorus concentration at week 28.
   
   Conclusion: In conclusion, our methodology and results provide a roadmap for assessing timely biomarker measures of exposures during specific tooth development to better understand the etiology of DDE for future prevention.

   @article{reed2024predictors,
     title = {Predictors of Developmental Defects of Enamel in Primary Maxillary Central Incisors Using Bayesian Model Selection},
     author = {Reed, Susan G and Fan, Sijian and Wagner, Carol L and Lawson, Andrew B},
     journal = {Caries research},
     volume = {58},
     number = {1},
     pages = {30--38},
     year = {2024},
     publisher = {S. Karger AG},
   }

2022

1. Clin. Imaging

   Prophylactic IVC Filter Placement in Patients with Severe Intracranial, Spinal Cord, and Orthopedic Injuries at High Thromboembolic Event Risk: A Utilization and Outcomes Analysis of the National Trauma Data Bank

   Scott J Lee, Sijian Fan, Mian Guo, Bill S Majdalany, and 5 more authors

   Clinical Imaging, 2022

   Abs Bib

   Purpose. To determine relationships between prophylactic inferior vena cava filter (IVCF) insertion and pulmonary embolism (PE), deep venous thrombosis (DVT), and in-hospital mortality outcomes in patients with severe traumatic pelvic/lower extremity, intracranial, and spinal cord injuries.
   
   Methods. Adult patients with severe traumatic pelvic/lower extremity, intracranial, and spinal cord injuries admitted to level I–IV trauma centers were selected from the National Trauma Data Bank (NTDB). IVCFs inserted within 48 hours after admission and before a lower extremity venous ultrasound were defined as prophylactic. Associations between prophylactic IVCF insertion and PE, DVT, and overall mortality were estimated using logistic regression models after propensity score matching.
   
   Results. Of 462,838 patients, 11,938 (2.6%) underwent prophylactic IVCF insertion. Prophylactic IVCF utilization decreased over time (6.3% in 2008 to 1.8% in 2015). Prophylactic IVCF placement was positively associated with PE (OR: 5.25, p < 0.01) and DVT (OR: 5.55, p < 0.01), but negatively associated with in-hospital mortality (OR: 0.46, p < 0.01).
   
   Conclusion. Prophylactic IVCF insertion was negatively associated with in-hospital mortality but positively associated with venous thromboembolism. Further research on prophylactic IVCF use in trauma patients with specific severe injury patterns is warranted.

   @article{lee2022prophylactic,
     title = {Prophylactic IVC Filter Placement in Patients with Severe Intracranial, Spinal Cord, and Orthopedic Injuries at High Thromboembolic Event Risk: A Utilization and Outcomes Analysis of the National Trauma Data Bank},
     author = {Lee, Scott J and Fan, Sijian and Guo, Mian and Majdalany, Bill S and Newsome, Janice and Duszak Jr, Richard and Gichoya, Judy and Benjamin, Elizabeth R and Kokabi, Nima},
     journal = {Clinical Imaging},
     volume = {91},
     pages = {134--140},
     year = {2022},
     publisher = {Elsevier},
   }

2. Clin. Imaging

   Comparative Effectiveness of Pelvic Arterial Embolization versus Laparotomy in Adults with Pelvic Injuries: A National Trauma Data Bank Analysis

   Abuzar Moradi Tuchayi, Nariman Nezami, Yuchen Zhang, Tarek N Hanna, and 7 more authors

   Clinical Imaging, 2022

   Abs Bib

   Purpose: To compare the clinical outcomes and trends of arterial embolization (AE) versus laparotomy used in the management of pelvic trauma.
   
   Materials and methods: Adult patients with pelvic injuries were identified using the National Trauma Data Bank (NTDB) from 2007 to 2015. Patients with non-pelvic life-threatening injuries were excluded. Patients were grouped into operatively managed pelvic ring injuries, laparotomy ± fixation, AE ± fixation, and laparotomy and AE ± fixation. Using linear mixed regression and logistic regression models, hospital length of stay (LOS), ICU days, ventilator days, and mortality across different therapies were compared. A propensity score weighting method was used to further eliminate treatment selection bias and compare outcomes between AE and laparotomy.
   
   Results: Of 7473 pelvic trauma patients, 1226 (16.4%) patients were only operatively managed. A total of 3730 patients (49.9%) underwent laparotomy, 2136 (28.6%) underwent AE, and 381 (5.1%) underwent both laparotomy and AE. The year of injury, patient age, gender, race, injury severity, and presence of shock were predictors of receipt of different therapies (P < 0.001 for all). After adjusting for these confounding factors, mortality was lower in the AE group compared with the laparotomy group (6.6% vs. 20.6%, P < 0.001). In addition, LOS and ICU days were shorter in the AE group than in the laparotomy group (P < 0.001).
   
   Conclusion: Arterial embolization in patients with pelvic injuries is associated with lower mortality, as well as shorter hospital length of stay and ICU stays compared to laparotomy.

   @article{tuchayi2022comparative,
     title = {Comparative Effectiveness of Pelvic Arterial Embolization versus Laparotomy in Adults with Pelvic Injuries: A National Trauma Data Bank Analysis},
     author = {Tuchayi, Abuzar Moradi and Nezami, Nariman and Zhang, Yuchen and Hanna, Tarek N and Johnson, Jamlik-Omari and Newsome, Janice and Fan, Sijian and Duszak Jr, Richard and Benjamin, Elizabeth R and Nguyen, Jonathan and others},
     journal = {Clinical Imaging},
     volume = {86},
     pages = {75--82},
     year = {2022},
     publisher = {Elsevier},
   }

2021

1. Acad. Radiol.

   Management of Splenic Trauma in Contemporary Clinical Practice: A National Trauma Data Bank Study

   Amanda H Chahine, Shenise Gilyard, Tarek N Hanna, Sijian Fan, and 6 more authors

   Academic Radiology, 2021

   Abs Bib

   Background: To evaluate the utilization and efficacy of various treatments for management of adult patients with splenic trauma, highlighting the evolving role of splenic artery embolization.
   
   Materials and Methods: The National Trauma Data Bank (NTDB) was queried for patients who sustained splenic trauma between 2007 and 2015, excluding those with death on arrival and selected nonsplenic high-grade injuries. Patients were categorized into nonoperative management (NOM), embolization, splenectomy, splenic repair, and combined treatment groups. Evaluated outcomes included hospital length of stay (LOS), intensive care unit LOS, mortality, and failures of NOM and embolization.
   
   Results: Overall, 117,743 patients with splenic-predominant trauma were included. Over the 9-year study period, 85,793 (72.9%) were treated with NOM, 21,999 (18.9%) with splenectomy, 3895 (3.3%) with embolization, and 2131 (1.8%) with splenic repair. From 2007 to 2015, mortality rates declined from 7.6% to 4.7%. The rate of NOM did not significantly change over time, while embolization increased by 369% (1.3% to 4.8%). Failure of NOM decreased from 4.4% in 2007 to 3.4% in 2015. Across all injury grades, NOM had the shortest LOS (8.3 days), followed by splenic repair (12.3 days), embolization (12.6 days), and splenectomy (13.8 days) (p < 0.001). After adjustment for clinical factors including injury severity, mortality rates were 7.1% for splenectomy, 3.2% for embolization, and 2.5% for NOM.
   
   Conclusion: Most patients with splenic-dominant blunt trauma are managed with nonoperative management. Over time, the use of embolization has increased while open surgery has declined, and mortality has improved across all treatment methods. Compared with splenectomy, embolization is associated with shorter hospital length of stay but remains relatively infrequently used.

   @article{chahine2021management,
     title = {Management of Splenic Trauma in Contemporary Clinical Practice: A National Trauma Data Bank Study},
     journal = {Academic Radiology},
     author = {Chahine, Amanda H and Gilyard, Shenise and Hanna, Tarek N and Fan, Sijian and Risk, Benjamin and Johnson, Jamlik Omari and Duszak Jr, Richard and Newsome, Janice and Xing, Minzhi and Kokabi, Nima},
     volume = {28},
     pages = {S138--S147},
     year = {2021},
     publisher = {Elsevier},
   }

2. JVIR

   Contemporary Management of Pediatric Blunt Splenic Trauma: A National Trauma Databank Analysis

   Kaitlin Shinn, Shenise Gilyard, Amanda Chahine, Sijian Fan, and 7 more authors

   Journal of Vascular and Interventional Radiology, 2021

   Abs Bib

   Purpose: To quantify changes in the management of pediatric patients with isolated splenic injury from 2007 to 2015.
   
   Materials and Methods: Patients under 18 years old with registered splenic injury in the National Trauma Data Bank (2007–2015) were identified. Splenic injuries were categorized into five management types: nonoperative management (NOM), embolization, splenic repair, splenectomy, or combination therapy. Linear mixed models accounting for confounding variables were used to examine the direct impact of management on length of stay (LOS), intensive care unit (ICU) days, and ventilator days.
   
   Results: Of included patients (n = 24,128), 90.3% (n = 21,789), 5.6% (n = 1,361), and 2.7% (n = 640) had NOM, splenectomy, and embolization, respectively. From 2007 to 2015, the rate of embolization increased from 1.5% to 3.5%, and the rate of splenectomy decreased from 6.9% to 4.4%. When combining injury grades, NOM was associated with the shortest LOS (5.1 days), ICU days (1.9 days), and ventilator days (0.5 day). Splenectomy was associated with longer LOS (10.1 days), ICU days (4.5 days), and ventilator days (2.1 days) compared with NOM. The average failure rate of NOM was 1.5% (180 failures out of 12,378 cases), and average embolization failure was 1.3% (6 failures out of 456 cases). Splenic artery embolization was associated with lower mortality than splenectomy (OR: 0.10, P < 0.001). No statistically significant difference in mortality was observed between embolization and NOM (OR: 0.96, P = 1.0).
   
   Conclusions: In pediatric splenic injury, nonoperative management is the most utilized approach and is associated with favorable outcomes, most notably in grades III to V injuries. When intervention is required, embolization is effective and increasingly utilized, particularly in lower grade injuries.

   @article{shinn2021contemporary,
     title = {Contemporary Management of Pediatric Blunt Splenic Trauma: A National Trauma Databank Analysis},
     author = {Shinn, Kaitlin and Gilyard, Shenise and Chahine, Amanda and Fan, Sijian and Risk, Benjamin and Hanna, Tarek and Johnson, Jamlik-Omari and Hawkins, C Matthew and Xing, Minzhi and Duszak Jr, Richard and others},
     journal = {Journal of Vascular and Interventional Radiology},
     volume = {32},
     number = {5},
     pages = {692--702},
     year = {2021},
     publisher = {Elsevier},
   }

2020

1. Thesis

   Improved Algorithm for Independent Component Analysis (ICA) with the Relax and Split Approximation

   Sijian Fan

   Emory University, 2020

   Abs Bib Code

   Independent component analysis (ICA) has been increasingly used to separate sources and extract features in signal processing and neuroimaging studies. To overcome its computational problems with local optima, as well as problems with non-smooth and non-convex objective functions, relax and split optimization was applied in this study and comparisons were made between the refined algorithms and the popular FastICA algorithm. A tuning parameter was used to control the relaxation and sparsity level of the Relax-Laplace method (with an objective function derived from the Laplace density), and to control the relaxation level of the Relax-logistic method (with an objective function derived from the logistic density).
   
   We conducted a simulation study to examine the impact of the tuning parameter on accuracy and sensitivity to initialization. We found that smaller values of the tuning parameter can lead to accurate estimates of the components while having fewer issues with local optima relative to FastICA, whereas larger values can result in inaccuracies. Running 1000 times with a pool of 50 initializations, we found that the Relax-Laplace algorithm was the most accurate and consistent compared with Relax-logistic, FastICA-logistic, and FastICA-tanh.
   
   We conducted a multi-subject analysis of functional magnetic resonance imaging (fMRI) data from the Human Connectome Project using Relax-Laplace, FastICA-logistic, and FastICA-tanh. In a pool of 50 initializations, the Relax-Laplace method returned the same result for all initializations, whereas both FastICA-logistic and FastICA-tanh converged to the argmax estimate in just over half of the initializations. Moreover, the Relax-Laplace produced sparse representations for the rs-fMRI data that highlight features of resting-state networks.

   @mastersthesis{fan2020improved,
     title = {Improved Algorithm for Independent Component Analysis (ICA) with the Relax and Split Approximation},
     author = {Fan, Sijian},
     school = {Emory University},
     year = {2020},
   }