Team name
TJU-Pichia
Pichia Codon DesignVirtual Cells for Protein ExpressionComparing Model-Generated and Rule-Based Coding Sequences
School of Synthetic Biology and Biomanufacturing, Tianjin UniversityT5 Virtual Cell
Project title
Pichia Codon DesignVirtual Cells for Protein Expression
T5 Virtual CellPichia Codon DesignGeneration Benchmarks
TJU-Pichia compares model-based and rule-based codon optimization with shared metrics, using first-round results to define constraints for the next design cycle.
- Team name
- TJU-Pichia
- Institution
- School of Synthetic Biology and Biomanufacturing, Tianjin University
Team media is for event showcases and reporting only. Portrait rights remain with the individuals. For concerns about portrait use, contact the organizers. We will investigate and take appropriate action.
Background
Host-Specific Codon Adaptation
Lactoferrin demand motivates production beyond limited natural extraction. Pichia is a common expression host, but its codon preferences differ from the native coding sequence. The project compares conventional high-frequency codon substitution with sequence generation that also considers length and expression constraints, aiming to adapt coding DNA while preserving the protein sequence.
Research Approach
Model and Rule Baselines
The first round feeds one lactoferrin protein sequence into Pichia-CLM without post-generation rule filtering and compares it with high-frequency codon substitution. Identical scripts calculate GC, CAI, ENC, rare-codon counts and composite scores; code and FASTA outputs are archived in GitLab. The team prerecords expectations of higher CAI and scores for rule optimization, versus shorter, less diverse model output.
- Target Host
- Pichia pastoris
- Compared Routes
- Model-Based vs Rule-Based
- Evaluation Measures
- GC、CAI、ENC
Research Process
Development & Validation
- 01
Stage 1
Comparing the Reported Outputs
Reported lengths are 498 bp for model output and 2,073 bp for rule output. GC is 33.53% versus 37.39%; CAI 0.8011 versus 0.905; host-preferred codons 0.6% versus 30.68%; and composite scores 31.92 versus 76.04. Both have zero counted rare codons. The model output meets the basic CAI threshold but fits host preferences less well; these sequence metrics do not establish equivalent full-length expression. - 02
Stage 2
Combining Generation with Constraints
The first round found shorter, lower-GC model output and better codon adaptation in rule-based output. The next planned pipeline combines model generation with multiple post-generation constraints. - 03
Stage 3
Cross-Disciplinary Model Tuning
Computational and experimental members initially struggled to align assumptions, and later measurements fell short of expectations. Joint discussions, problem review and divided troubleshooting helped refine the workflow.
Project Evidence
Methods & Figures
Teamwork
Computational, experimental and candidate review records
Team films
