Bumgardner, C. | Search bioRxiv

bioRxiv2025

Systematic contextual biases in SegmentNT relevant to all nucleotide transformer models

Recent advances in large language models (LLMs) have extended to genomic applications, yet model robustness relative to context is unclear. Here, we demonstrate two intrinsic biases (input sequence length and nucleotide position) affecting SegmentNT results, a model included with the Nucleotide Transformer that provides nucleotide-level predictions of biological features. We demonstrate that nucleotide position within the input sequence (beginning, middle, or end) alters the nature of SegmentNTs raw prediction probabilities, which can be standardized to improve prediction consistency. While longer input sequence length improves model performance, diminishing returns suggest a surprisingly small input length of [~]3,072 nucleotides might be sufficient for many applications. We further identify a 24-nucleotide periodic oscillation in SegmentNTs prediction probabilities, revealing an intrinsic bias potentially linked to the models training tokenization (6-mers) and architecture. We identify potential approaches to account for these biases and provide generalizable insights for utilizing nucleotide-resolution functional prediction models.

bioinformatics↗

bioRxiv2024

Behavioral and genetic markers of susceptibility to escalate fentanyl intake.

BackgroundThe "loss of control" over drug consumption, present in opioid use disorder (OUD) and known as escalation of intake, is well-established in preclinical rodent models. However, little is known about how antecedent behavioral characteristics, such as valuation of hedonic reinforcers prior to drug use, may impact the trajectory of fentanyl intake over time. Moreover, it is unclear if distinct escalation phenotypes may be driven by genetic markers predictive of OUD susceptibility. MethodsMale and female Sprague-Dawley rats (n=63) were trained in a sucrose reinforcement task using a progressive ratio schedule. Individual differences in responsivity to sucrose were hypothesized to predict escalation of fentanyl intake. Rats underwent daily 1-h acquisition sessions for i.v. fentanyl self-administration (2.5 {micro}g/kg; FR1) for 7 days, followed by 21 6-h escalation sessions, then tissue from prefrontal cortex was collected for RNA sequencing and qPCR. Latent growth curve and group-based trajectory modeling were used, respectively, to evaluate the association between sucrose reinforcement and fentanyl self-administration and to identify whether distinct escalation phenotypes can be linked to gene expression patterns. ResultsSucrose breakpoints were not predictive of fentanyl acquisition nor change during escalation, but did predict fentanyl intake on the first day of extended access to fentanyl. Permutation analyses did not identify associations between behavior and single gene expression when evaluated overall, or between our ascertained phenotypes. However, weighted genome correlation network analysis (WGCNA) and gene set enrichment analysis (GSEA) determined several gene modules linked to escalated fentanyl intake, including genes coding for voltage-gated potassium channels, calcium channels, and genes involved in excitatory synaptic signaling. Transcription factor analyses identified EZH2 and JARID2 as potential transcriptional regulators associated with escalated fentanyl intake. Genome-wide association study (GWAS) term categories were also generated and positively associated with terms relating to substance use disorders. DiscussionEscalation of opioid intake is largely distinct from motivation for natural reward, such as sucrose. Further, the gene networks associated with fentanyl escalation suggest that engagement of select molecular pathways distinguish individuals with "addiction prone" behavioral endophenotypes, potentially representing druggable targets for opioid use disorder. Our extended in silico identification of SNPs and transcription factors associated with the "addiction prone" high escalating rats highlights the importance of integrating findings from translational preclinical models. Through a precision medicine approach, our results may aid in the development of patient-centered treatment options for those with OUD.

neuroscience↗