Biology subjects

Choi, J. D.

Publications and source records attributed to Choi, J. D..

4 recordsLinked to original sources

bioRxiv2026

Adoption of MMPose, a general purpose pose estimation library, for animal tracking

1Markerless pose estimation has emerged as a powerful technique for animal behavior quantification, capable of high resolution tracking of body parts. Many neuroscience labs rely on tools like DeepLabCut and SLEAP, which provide accessible interfaces but restrict users to a narrow set of models and configurations. In this work, we adopt MMPose an open source, general-purpose computer vision library to build a workflow for training and evaluating multiple state-of-the-art models on animal video datasets. We benchmark these models in two scenarios: (1) a complex maze assay with occlusions and varied backgrounds, and (2) a simpler open field arena with a high-contrast background. Our results show that a bottomup model (DEKR) delivers the highest accuracy in the complex task, whereas lighter-weight models (e.g., SLEAP) offer superior speed highlighting a clear trade-off between accuracy and throughput. We also evaluate a recently published foundation model (TopViewMouse-5K) trained on a large top-view mouse dataset to test its generalization. It performs poorly on our tasks at zero-shot, and even when we combine its data with our training set, we observe no consistent benefit. These findings emphasize the importance of context-specific model selection and the need for more diverse training data to create generalizable pose models. By leveraging a general-purpose vision library, researchers can flexibly choose models that best suit their experimental needs. This work illustrates how adopting advanced computer vision frameworks can accelerate behavioral neuroscience and genetics research, paving the way for more scalable, reproducible, and sensitive analysis of animal behavior.

neuroscience↗

bioRxiv2026

Optimizing Intermediate Representations: A Framework for Low-Cost, High-Accuracy Behavior Quantification

Quantitative measurement of animal behavior is a cornerstone of neuroscience, genetics, and ethology. While modern computer vision has democratized automated analysis, the field has coalesced around pose estimation as the standard intermediate representation. This reliance imposes a significant bottleneck: researchers must often train custom pose models using large, labor-intensive datasets. Furthermore, the assumption that denser anatomical tracking yields better classification remains largely unverified. Here, we benchmark intermediate representations for supervised mouse behavior classification to determine the optimal trade-off between annotation cost and model performance. We systematically evaluate the sensitivity of classification to keypoint density, the impact of temporal feature engineering, and the viability of segmentation-derived shape descriptors as a low-cost alternative. We find that classifier performance is remarkably robust to keypoint variation; increasing keypoint density yields negligible gains, particularly when behavior training sets are sufficiently large. In contrast, augmenting models with temporal features (specifically FFT-based signal processing) consistently drives performance improvements. Crucially, we demonstrate that whole-body segmentation achieves performance parity with explicit pose estimation across most behaviors. These findings challenge the "more is better" intuition in pose tracking and suggest a paradigm shift: efficient pipelines should prioritize behavioral dataset volume and temporal dynamics over complex anatomical keypoints.

neuroscience↗

bioRxiv2020

SINE Retrotransposons Import Polyadenylation Signals to 3'UTRs in Dog (Canis familiaris)

BackgroundMessenger RNA 3 untranslated regions (3UTRs) control many aspects of gene expression and determine where the transcript will terminate. The polyadenylation signal (PAS) AAUAAA is a key regulator of transcript termination and this hexamer, or a similar sequence, is very frequently found within 30 bp of 3UTR ends. Short interspersed element (SINE) retrotransposons are found throughout genomes in high copy number. When inserted into genes they can disrupt expression, alter splicing, or cause nuclear retention of mRNAs. The genomes of the domestic dog and other carnivores carry hundreds of thousands Can-SINEs, a tRNA-related SINE with transcription termination potential. Because of this we asked whether Can-SINEs may help terminate transcript in some dog genes. ResultsDog 3UTRs have several peaks of AATAAA PAS frequency within 40 bp of the 3UTR end, including four bp-interval peaks at 28, 32, and 36 bp from the end. The periodicity is partly explained by TAAA(n) repeats within Can-SINE AT-rich tails. While density of antisense-oriented Can-SINEs in 3UTRs is fairly constant with distances from 3end, sense-oriented Can-SINEs are common at the 3end but nearly absent farther upstream. There are nine Can-SINE sub-types in the dog genome and the consensus sequence sense strands (head to tail) all carry at least three PASs while antisense strands usually have none. We annotated all repeat-masked Can-SINE copies in the Boxer reference genome and found that the young SINEC_Cf type has a mode of 15 bp for target site duplications (TSDs). We find that all Can-SINE types favor integration at TSDs beginning with A(4). The count of AATAAA PASs differs significantly between sense and antisense-oriented retrotransposons in transcripts. Can-SINEs near 3UTR ends are very likely to carry AATAAA on the mRNA sense strand while those farther upstream are not. We also identified loci where Can-SINE insertion has truncated or altered a dog 3UTR compared to the human ortholog. ConclusionDog Can-SINE activity has imported AATAAA PASs into gene transcripts and led to alteration of 3UTRs. AATAAA sequences are selectively removed from Can-SINEs in introns and upstream 3UTR regions but are retained at the far downstream end of 3UTRs, which we infer reflects their role as termination sequences for these transcripts.

genomics↗

bioRxiv2020

Polymorphic SINEC_Cf Retrotransposons in the Genome of the Dog (Canis familiaris)

The dog is an exciting genetic system in which many simple and complex traits have now been mapped. For many traits the causal mutation is a polymorphic SINE. To investigate the genome-wide pattern of young SINEC_Cf insertions, we sampled 62 dogs representing 59 breeds and sequenced libraries enriched for SINE flanks. In each dog we detect an average of 10,423 polymorphic loci and all together the libraries identify 81,747 putative polymorphic SINEs. We validated 184 SINEs inserted in protein-coding exons, untranslated regions, introns and intergenic sequence. In dogs both SINEC_Cf and LINEs exhibit a strand bias in introns where antisense copies are more frequent. Antisense polymorphic SINEs also have a higher density in introns. Both SINEs and LINEs drop to very low density near exons. Both sense and antisense polymorphic SINEs also drop to low density upstream of coding exons but not downstream. Antisense polymorphic SINEC_Cfs upstream of coding exons are known to cause narcolepsy, merle, and progressive retinal atrophy in dogs. In other mammals SINE pairs in inverted orientation disrupt gene expression. We find inverted pairs of SINEC_Cf are rare in both introns and intergenic sequence when the two SINEs are separated by less than 100 bp. The lack of inverted pairs is even more pronounced when the SINEs have high sequence identity. Intronic and intergenic LINE pairs show similar patterns. Polymorphic SINEs rarely pair with either SINEC_Cf or SINEC_Cf2. Overall, the high insertion rate of SINEC_Cf provides a natural mutagenesis screen in the dog genome.

genomics↗