Chun Fan
Chun Fan has a total of 111 co-authors. Here are the top co-authors by number of shared publications:
- Anders M. Dale
Multimodal Imaging Laboratory, University of California, San Diego, La Jolla, USA
5 shared publications - Ole Andreas Andreassen
Institute of Clinical Medicine, University of Oslo, Oslo, Norway
5 shared publications - Oleksandr Frei
Centre for Precision Psychiatry, Division of Mental Health and Addiction, Oslo University Hospital and Institute of Clinical Medicine, University of Oslo, Oslo, Norway
3 shared publications - Wesley K. Thompson
Population Neuroscience and Genetics (PoNG) Center, Laureate Institute for Brain Research, Tulsa, OK, USA
3 shared publications - Xiaofei Sun
School of Petroleum Engineering, China University of Petroleum (East China), Qingdao 266580, China
3 shared publications - Carolina Makowski
University of California San Diego (UCSD) La Jolla California USA
2 shared publications
Latest Publications
- 23 September 2022
- Published by Elsevier in European Neuropsychopharmacology
- Vol. 63, e22-E22
AGE-DEPENDENT GENOME-WIDE ASSOCIATION STUDY OF STRUCTURAL BRAIN MORPHOLOGY IN ABCD
- 23 September 2022
- Published by Elsevier in European Neuropsychopharmacology
- Vol. 63, e22-E22
AGE-DEPENDENT GENOME-WIDE ASSOCIATION STUDY OF STRUCTURAL BRAIN MORPHOLOGY IN ABCD
- 23 September 2022
- Published by Elsevier in European Neuropsychopharmacology
- Vol. 63, e56-e57
22. ABUNDANT PLEIOTROPY ACROSS NEUROIMAGING MODALITIES IDENTIFIED THROUGH MULTIVARIATE GENOME-WIDE ASSOCIATION STUDIES
- 23 September 2022
- Published by Elsevier in European Neuropsychopharmacology
- Vol. 63, e56-e57
22. ABUNDANT PLEIOTROPY ACROSS NEUROIMAGING MODALITIES IDENTIFIED THROUGH MULTIVARIATE GENOME-WIDE ASSOCIATION STUDIES
- Open Access
Summarize, Outline, and Elaborate: Long-Text Generation via Hierarchical Supervision from Extractive Summaries
The difficulty of generating coherent long texts lies in the fact that existing models overwhelmingly focus on predicting local words, and cannot make high level plans on what to generate or capture the high-level discourse dependencies between chunks of texts. Inspired by human writing processes, where a list of bullet points or a catalog is first outlined, and then each bullet point is expanded to form the whole article, we propose {\it SOE}, a pipelined system that involves of summarizing, outlining and elaborating for long text generation: the model first outlines the summaries for different segments of long texts, and then elaborates on each bullet point to generate the corresponding segment. To avoid the labor-intensive process of summary soliciting, we propose the {\it reconstruction} strategy, which extracts segment summaries in an unsupervised manner by selecting its most informative part to reconstruct the segment. The proposed generation system comes with the following merits: (1) the summary provides high-level guidance for text generation and avoids the local minimum of individual word predictions; (2) the high-level discourse dependencies are captured in the conditional dependencies between summaries and are preserved during the summary expansion process and (3) additionally, we are able to consider significantly more contexts by representing contexts as concise summaries. Extensive experiments demonstrate that SOE produces long texts with significantly better quality, along with faster convergence speed.
Open AccessSummarize, Outline, and Elaborate: Long-Text Generation via Hierarchical Supervision from Extractive Summaries
The difficulty of generating coherent long texts lies in the fact that existing models overwhelmingly focus on predicting local words, and cannot make high level plans on what to generate or capture the high-level discourse dependencies between chunks of texts. Inspired by human writing processes, where a list of bullet points or a catalog is first outlined, and then each bullet point is expanded to form the whole article, we propose {\it SOE}, a pipelined system that involves of summarizing, outlining and elaborating for long text generation: the model first outlines the summaries for different segments of long texts, and then elaborates on each bullet point to generate the corresponding segment. To avoid the labor-intensive process of summary soliciting, we propose the {\it reconstruction} strategy, which extracts segment summaries in an unsupervised manner by selecting its most informative part to reconstruct the segment. The proposed generation system comes with the following merits: (1) the summary provides high-level guidance for text generation and avoids the local minimum of individual word predictions; (2) the high-level discourse dependencies are captured in the conditional dependencies between summaries and are preserved during the summary expansion process and (3) additionally, we are able to consider significantly more contexts by representing contexts as concise summaries. Extensive experiments demonstrate that SOE produces long texts with significantly better quality, along with faster convergence speed.
- Open Access
- 15 June 2022
- Published by American Association for Cancer Research (AACR) in Cancer Research
- Vol. 82 (12_Supplem) , 3825
Abstract 3825: Germline modifiers of the tumor immune microenvironment reveal drivers of immunotherapy response
With the continued promise of immunotherapy as an avenue for treating cancer, understanding how host genetics contributes to the tumor immune microenvironment (TIME) is essential to tailoring cancer risk screening and treatment strategies. Using genotypes from over 8,000 European individuals in The Cancer Genome Atlas and 137 heritable tumor immune phenotype components (IP components), we identified and investigated 532 TIME-SNPs. Focusing on 77 variants that were relevant to cancer risk, survival, or treatment response, we explored their potential to reveal novel targets for immunotherapy. Many variants overlapped regions with histone marks indicating active transcription, and influenced gene activities in specific immune cell subsets, such as macrophages and dendritic cells. TIME-SNPs implicated genes such as LAIR1, TREX1, CTSS, CTSW and LILRB2 were differentially expressed between responders and non-responders to immune-checkpoint blockade (ICB) in preclinical studies. Of these, LILRB2 and LAIR1 have already been identified as putative targets for immunotherapy. Here we found that inhibition of CTSS led to better tumor control and survival in murine models, alone or in combination with anti-PD-1. Collectively we show that through an integrative approach, it is possible to link host genetics to TIME characteristics, informing novel biomarkers for cancer risk and target identification in immunotherapy. Citation Format: Meghana Pagadala, Victoria Wu, Eva Pérez-Guijarro, Hyo Kim, Andrea Castro, James Talwar, Cristian Gonzalez-Colin, Steven Cao, Benjamin J. Schmiedel, Timothy Sears, Shervin Goudarzi, Divya Kirani, Rany M. Salem, Gerald P. Morris, Olivier Harismendy, Sandip P. Patel, Jill P. Mesirov, Maurizio Zanetti, Chi-Ping Day, Chun C. Fan, Wesley K. Thompson, Glenn Merlino, J. Silvio Gutkind, Pandurangan Vijayanand, Hannah Carter. Germline modifiers of the tumor immune microenvironment reveal drivers of immunotherapy response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3825.
- 15 June 2022
- Published by American Association for Cancer Research (AACR) in Cancer Research
- Vol. 82 (12_Supplem) , 3825
Open AccessAbstract 3825: Germline modifiers of the tumor immune microenvironment reveal drivers of immunotherapy response
With the continued promise of immunotherapy as an avenue for treating cancer, understanding how host genetics contributes to the tumor immune microenvironment (TIME) is essential to tailoring cancer risk screening and treatment strategies. Using genotypes from over 8,000 European individuals in The Cancer Genome Atlas and 137 heritable tumor immune phenotype components (IP components), we identified and investigated 532 TIME-SNPs. Focusing on 77 variants that were relevant to cancer risk, survival, or treatment response, we explored their potential to reveal novel targets for immunotherapy. Many variants overlapped regions with histone marks indicating active transcription, and influenced gene activities in specific immune cell subsets, such as macrophages and dendritic cells. TIME-SNPs implicated genes such as LAIR1, TREX1, CTSS, CTSW and LILRB2 were differentially expressed between responders and non-responders to immune-checkpoint blockade (ICB) in preclinical studies. Of these, LILRB2 and LAIR1 have already been identified as putative targets for immunotherapy. Here we found that inhibition of CTSS led to better tumor control and survival in murine models, alone or in combination with anti-PD-1. Collectively we show that through an integrative approach, it is possible to link host genetics to TIME characteristics, informing novel biomarkers for cancer risk and target identification in immunotherapy. Citation Format: Meghana Pagadala, Victoria Wu, Eva Pérez-Guijarro, Hyo Kim, Andrea Castro, James Talwar, Cristian Gonzalez-Colin, Steven Cao, Benjamin J. Schmiedel, Timothy Sears, Shervin Goudarzi, Divya Kirani, Rany M. Salem, Gerald P. Morris, Olivier Harismendy, Sandip P. Patel, Jill P. Mesirov, Maurizio Zanetti, Chi-Ping Day, Chun C. Fan, Wesley K. Thompson, Glenn Merlino, J. Silvio Gutkind, Pandurangan Vijayanand, Hannah Carter. Germline modifiers of the tumor immune microenvironment reveal drivers of immunotherapy response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3825.
- 28 April 2022
- Published by Elsevier in Biological Psychiatry
- Vol. 91 (9) , S76-S77
Clustering of Genetic Effects on Cortical Morphology and Links to Psychiatric Disorders
- 28 April 2022
- Published by Elsevier in Biological Psychiatry
- Vol. 91 (9) , S76-S77
Clustering of Genetic Effects on Cortical Morphology and Links to Psychiatric Disorders
- Open Access
$k$Folden: $k$-Fold Ensemble for Out-Of-Distribution Detection
Out-of-Distribution (OOD) detection is an important problem in natural language processing (NLP). In this work, we propose a simple yet effective framework $k$Folden, which mimics the behaviors of OOD detection during training without the use of any external data. For a task with $k$ training labels, $k$Folden induces $k$ sub-models, each of which is trained on a subset with $k-1$ categories with the left category masked unknown to the sub-model. Exposing an unknown label to the sub-model during training, the model is encouraged to learn to equally attribute the probability to the seen $k-1$ labels for the unknown label, enabling this framework to simultaneously resolve in- and out-distribution examples in a natural way via OOD simulations. Taking text classification as an archetype, we develop benchmarks for OOD detection using existing text classification datasets. By conducting comprehensive comparisons and analyses on the developed benchmarks, we demonstrate the superiority of $k$Folden against current methods in terms of improving OOD detection performances while maintaining improved in-domain classification accuracy. The code and datasets can be found at: \url{https://github.com/ShannonAI/kfolden-ood-detection}.Open Access$k$Folden: $k$-Fold Ensemble for Out-Of-Distribution Detection
Out-of-Distribution (OOD) detection is an important problem in natural language processing (NLP). In this work, we propose a simple yet effective framework $k$Folden, which mimics the behaviors of OOD detection during training without the use of any external data. For a task with $k$ training labels, $k$Folden induces $k$ sub-models, each of which is trained on a subset with $k-1$ categories with the left category masked unknown to the sub-model. Exposing an unknown label to the sub-model during training, the model is encouraged to learn to equally attribute the probability to the seen $k-1$ labels for the unknown label, enabling this framework to simultaneously resolve in- and out-distribution examples in a natural way via OOD simulations. Taking text classification as an archetype, we develop benchmarks for OOD detection using existing text classification datasets. By conducting comprehensive comparisons and analyses on the developed benchmarks, we demonstrate the superiority of $k$Folden against current methods in terms of improving OOD detection performances while maintaining improved in-domain classification accuracy. The code and datasets can be found at: \url{https://github.com/ShannonAI/kfolden-ood-detection}. - Open Access
- 28 October 2021
- Published by Cold Spring Harbor Laboratory in bioRxiv
FEMA: Fast and efficient mixed-effects algorithm for large sample whole-brain imaging data
The linear mixed-effects model (LME) is a versatile approach to account for dependence among observations. Many large-scale neuroimaging datasets with complex designs have increased the need for LME, however LME has seldom been used in whole-brain imaging analyses due to its heavy computational requirements. In this paper, we introduce a fast and efficient mixed-effects algorithm (FEMA) that makes whole-brain vertex-wise, voxel-wise, and connectome-wide LME analyses in large samples possible. We validate FEMA with extensive simulations, showing that the estimates of the fixed effects are equivalent to standard maximum likelihood estimates but obtained with orders of magnitude improvement in computational speed. We demonstrate the applicability of FEMA by studying the cross-sectional and longitudinal effects of age on region-of-interest level and vertex-wise cortical thickness, as well as connectome-wide functional connectivity values derived from resting state functional MRI, using longitudinal imaging data from the Adolescent Brain Cognitive DevelopmentSMStudy release 4.0. Our analyses reveal distinct spatial patterns for the annualized changes in vertex-wise cortical thickness and connectome-wide connectivity values in early adolescence, highlighting a critical time of brain maturation. The simulations and application to real data show that FEMA enables advanced investigation of the relationships between large numbers of neuroimaging metrics and variables of interest while considering complex study designs, including repeated measures and family structures, in a fast and efficient manner. The source code for FEMA is available via:https://github.com/cmig-research-group/cmig_tools/.
- 28 October 2021
- Published by Cold Spring Harbor Laboratory in bioRxiv
Open AccessFEMA: Fast and efficient mixed-effects algorithm for large sample whole-brain imaging data
The linear mixed-effects model (LME) is a versatile approach to account for dependence among observations. Many large-scale neuroimaging datasets with complex designs have increased the need for LME, however LME has seldom been used in whole-brain imaging analyses due to its heavy computational requirements. In this paper, we introduce a fast and efficient mixed-effects algorithm (FEMA) that makes whole-brain vertex-wise, voxel-wise, and connectome-wide LME analyses in large samples possible. We validate FEMA with extensive simulations, showing that the estimates of the fixed effects are equivalent to standard maximum likelihood estimates but obtained with orders of magnitude improvement in computational speed. We demonstrate the applicability of FEMA by studying the cross-sectional and longitudinal effects of age on region-of-interest level and vertex-wise cortical thickness, as well as connectome-wide functional connectivity values derived from resting state functional MRI, using longitudinal imaging data from the Adolescent Brain Cognitive DevelopmentSMStudy release 4.0. Our analyses reveal distinct spatial patterns for the annualized changes in vertex-wise cortical thickness and connectome-wide connectivity values in early adolescence, highlighting a critical time of brain maturation. The simulations and application to real data show that FEMA enables advanced investigation of the relationships between large numbers of neuroimaging metrics and variables of interest while considering complex study designs, including repeated measures and family structures, in a fast and efficient manner. The source code for FEMA is available via:https://github.com/cmig-research-group/cmig_tools/.
- Open Access
- 1 January 2021
- Published by Association for Computational Linguistics (ACL)
- p. 3079-3090
Layer-wise Model Pruning based on Mutual Information
Chun Fan, Jiwei Li, Tianwei Zhang, Xiang Ao, Fei Wu, Yuxian Meng, Xiaofei Sun. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.
- 1 January 2021
- Published by Association for Computational Linguistics (ACL)
- p. 3079-3090
Open AccessLayer-wise Model Pruning based on Mutual Information
Chun Fan, Jiwei Li, Tianwei Zhang, Xiang Ao, Fei Wu, Yuxian Meng, Xiaofei Sun. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.
- Open Access
Personalized genetic assessment of age associated Alzheimer’s disease risk
Importance: Identifying individuals at risk for developing Alzheimer’s disease (AD) is of utmost importance. Although genetic studies have identified APOE and other AD associated single nucleotide polymorphisms (SNPs), genetic information has not been integrated into an epidemiological framework for personalized risk prediction. Objective: To develop, replicate and validate a novel polygenic hazard score for predicting age-specific risk for AD. Setting: Multi-center, multi-cohort genetic and clinical data. Participants: We assessed genetic data from 17,008 AD patients and 37,154 controls from the International Genetics of Alzheimer’s Project (IGAP), and 6,409 AD patients and 9,386 older controls from Phase 1 Alzheimer’s Disease Genetics Consortium (ADGC). As independent replication and validation cohorts, we also evaluated genetic, neuroimaging, neuropathologic, CSF and clinical data from ADGC Phase 2, National Institute of Aging Alzheimer’s Disease Center (NIA ADC) and Alzheimer’s Disease Neuroimaging Initiative (ADNI) (total n = 20,680) Main Outcome(s) and Measure(s): Use the IGAP cohort to first identify AD associated SNPs (at p < 10 -5 ). Next, integrate these AD associated SNPs into a Cox proportional hazards model using ADGC phase 1 genetic data, providing a polygenic hazard score (PHS) for each participant. Combine population based incidence rates, and genotype-derived PHS for each individual to derive estimates of instantaneous risk for developing AD, based on genotype and age. Finally, assess replication and validation of PHS in independent cohorts. Results: Individuals in the highest PHS quantiles developed AD at a considerably lower age and had the highest yearly AD incidence rate. Among APOE ε3/3 individuals, PHS modified expected age of AD onset by more than 10 years between the lowest and highest deciles. In independent cohorts, PHS strongly predicted empirical age of AD onset (p = 1.1 x 10 -26 ), longitudinal progression from normal aging to AD (p = 1.54 x 10 -10 ) and associated with markers of AD neurodegeneration. Conclusions: We developed, replicated and validated a clinically usable PHS for quantifying individual differences in age-specific risk of AD. Beyond APOE , polygenic architecture plays an important role in modifying AD risk. Precise quantification of AD genetic risk will be useful for early diagnosis and therapeutic strategies.
Open AccessPersonalized genetic assessment of age associated Alzheimer’s disease risk
Importance: Identifying individuals at risk for developing Alzheimer’s disease (AD) is of utmost importance. Although genetic studies have identified APOE and other AD associated single nucleotide polymorphisms (SNPs), genetic information has not been integrated into an epidemiological framework for personalized risk prediction. Objective: To develop, replicate and validate a novel polygenic hazard score for predicting age-specific risk for AD. Setting: Multi-center, multi-cohort genetic and clinical data. Participants: We assessed genetic data from 17,008 AD patients and 37,154 controls from the International Genetics of Alzheimer’s Project (IGAP), and 6,409 AD patients and 9,386 older controls from Phase 1 Alzheimer’s Disease Genetics Consortium (ADGC). As independent replication and validation cohorts, we also evaluated genetic, neuroimaging, neuropathologic, CSF and clinical data from ADGC Phase 2, National Institute of Aging Alzheimer’s Disease Center (NIA ADC) and Alzheimer’s Disease Neuroimaging Initiative (ADNI) (total n = 20,680) Main Outcome(s) and Measure(s): Use the IGAP cohort to first identify AD associated SNPs (at p < 10 -5 ). Next, integrate these AD associated SNPs into a Cox proportional hazards model using ADGC phase 1 genetic data, providing a polygenic hazard score (PHS) for each participant. Combine population based incidence rates, and genotype-derived PHS for each individual to derive estimates of instantaneous risk for developing AD, based on genotype and age. Finally, assess replication and validation of PHS in independent cohorts. Results: Individuals in the highest PHS quantiles developed AD at a considerably lower age and had the highest yearly AD incidence rate. Among APOE ε3/3 individuals, PHS modified expected age of AD onset by more than 10 years between the lowest and highest deciles. In independent cohorts, PHS strongly predicted empirical age of AD onset (p = 1.1 x 10 -26 ), longitudinal progression from normal aging to AD (p = 1.54 x 10 -10 ) and associated with markers of AD neurodegeneration. Conclusions: We developed, replicated and validated a clinically usable PHS for quantifying individual differences in age-specific risk of AD. Beyond APOE , polygenic architecture plays an important role in modifying AD risk. Precise quantification of AD genetic risk will be useful for early diagnosis and therapeutic strategies.
- 1 January 2010
- Published by IOP Publishing in Chinese Physics B
- Vol. 19 (1) , 014601-5
Valence band variation in Si (110) nanowire induced by a covered insulator
- 1 January 2010
- Published by IOP Publishing in Chinese Physics B
- Vol. 19 (1) , 014601-5
Valence band variation in Si (110) nanowire induced by a covered insulator