michaelml.dev / Michael_Portfolio.ipynb
ML pipeline for predicting RNA 3D structures. Used transformer architecture with geometric deep learning for molecular modeling.
[ ]
code · RNA 3D Folding Pipeline
project = load_project("rna-3d-folding")
project.summary()Output
RNA 3D Folding Pipeline
ML pipeline for predicting RNA 3D structures. Used transformer architecture with geometric deep learning for molecular modeling.
BioinformaticsTransformersTensorFlowMolecular MLGPU ComputingNumPyPandasData AugmentationGeometric DLStructural BiologySciPy
[ ]
code · Problem definition
project = load_project("rna-3d-folding")
project.problem()Output
ProblemPredict RNA three-dimensional coordinates from sequence and evolutionary context without direct structural measurement.
ApproachCombine sequence, pairing, MSA, and geometry features in a dual-tower Transformer with cross-attention and rotation augmentation.
OutcomeDocumented an end-to-end experimental RNA structure pipeline; no benchmark or deployment metric is claimed.
[ ]
code · Constraints
project = load_project("rna-3d-folding")
project.constraints()Output
Constraints
- RNA sequences vary in length and have complex long-range relationships.
- Predictions must respect three-dimensional geometry and rotation invariance.
- Evolutionary MSA information may be sparse or expensive to prepare.
[ ]
code · Role and implementation
project = load_project("rna-3d-folding")
project.implementation()Output
Role and implementation
My role: Built the preprocessing and modeling pipeline, including FASTA/MSA handling, geometric augmentation, dual encoders, and coordinate prediction.
- Encoded sequence, secondary-structure heuristics, conservation, and covariance features.
- Applied random 3D rotations to structural targets during training.
- Fused biology and geometry towers through cross-attention.
[ ]
code · Architecture
project = load_project("rna-3d-folding")
project.show_architecture()Output
Data IngestionLoad RNA sequences from FASTA files and parse target 3D coordinates
Feature EngineeringGenerate one-hot encodings, position-specific features, and process MSA data
Geometric ProcessingNormalize coordinates and apply data augmentation (rotation/translation)
Dual-Tower EncodingProcess inputs through parallel Biology and Geometry Transformer towers
Structure PredictionPredict final 3D coordinates using cross-attention and dense layers
[ ]
code · Demo and media
project = load_project("rna-3d-folding")
project.media()Output
[ ]
code · Evaluation
project = load_project("rna-3d-folding")
project.evaluation()Output
Evaluation
- Validated the data and model path as an experimental Kaggle notebook.
- No headline competition metric is present in the project record.
[ ]
code · Results
project = load_project("rna-3d-folding")
project.results()Output
Results
- Produced sequence-to-coordinate predictions through a reproducible experimental pipeline.
- The result is presented as a modeling study rather than a validated scientific tool.
[ ]
code · Tradeoffs and failures
project = load_project("rna-3d-folding")
project.tradeoffs()Output
Tradeoffs
- Dual towers preserve modality structure but require more memory and coordination than one encoder.
- Heuristic pairing features are practical but do not replace experimentally derived structure.
[ ]
code · What I would improve
project = load_project("rna-3d-folding")
project.improvements()Output
Improvements
- Add established RNA structure benchmarks and uncertainty reporting.
- Evaluate equivariant architectures and stronger geometric losses.
[ ]