You are browsing as a guest. Sign up (or log in) to start making projects!

MINTS

By @ArjunCodess

  • 6 Devlogs
  • 3 Total hours

MINTS is a reproducible mechanistic-interpretability pipeline for genomic transformers. It loads DNABERT-2 and Nucleotide Transformer backends, extracts QK/OV circuit matrices, probes frozen residual streams, scores CTCF motif support from JASPAR, tests QK-to-motif alignment and matched attention enrichment, runs custom DNABERT forward-hook activation patching, and searches for distributed CTCF-aligned SAE features.

Source
Open comments for this post

@ArjunCodess on MINTS

42m 21s logged

  • ran the full pipeline without any caps and regenerated all results.
  • updated the interpretations based on the new results.
  • updated the README and paper to match the final outputs.
  • completed the project (for now) and verified that the code, results, and documentation are all consistent.

image: source control image showing the main changes i did in the paper’s latex.

0
0
6
Ship #1 Changes requested

@ArjunCodess on MINTS

Built a reproducible mechanistic interpretability pipeline for genomic transformers. It extracts model circuits, runs activation patching, and tests biological motif detection.

The biggest challenge was building reliable experiments without overclaiming results. I’m proud of the scientific rigor, reproducibility, and honest negative findings.

Run python main.py to test everything.

  • 5 devlogs
  • 3h
Try project → See source code →
Open comments for this post

@ArjunCodess on MINTS

40m 4s logged

  • added frozen DNABERT sequence-head evaluation with AUROC, AUPRC, and accuracy reporting.
  • added real null calibration, GC-matched background calibration, and improved the threshold sensitivity analysis.
  • regenerated the downstream task performance and threshold sensitivity results.

image: source control image showing the main changes i did in the threshold sensitivity code.

0
0
7
Open comments for this post

@ArjunCodess on MINTS

22m 10s logged

  • replaced the evidence summary with a clearer evidence table linking each claim to its supporting results and limitations.
  • added a section explaining the main lessons for mechanistic interpretability.
  • expanded the limitations section with clearer discussion of the study’s scope and assumptions.

image: source control image showing the main changes i did in the paper.

0
0
12
Open comments for this post

@ArjunCodess on MINTS

18m 3s logged

  • added a glossary for the main biology terms.
  • explained why the selected motifs are biologically important.
  • simplified the theory section.
  • moved detailed proofs to the appendix.
  • clearly defined how nucleotide motifs are mapped to tokens.
  • added a simple workflow diagram for the full analysis pipeline.

image: source control image showing changes that i made in the code.

0
0
24
Open comments for this post

@ArjunCodess on MINTS

20m logged

made the entire pipeline fully reproducible, without any caps, including the threshold analysis and result generation.

image: screenshot of the pipeline being run.

0
0
3
Open comments for this post

@ArjunCodess on MINTS

58m 16s logged

  • showed model performance before the interpretability analysis.
  • separated task performance from probe results.
  • combined all interpretability analyses into one pipeline.
  • made CTCF the main case study.
  • made the overall framing clearer.
  • explained every analysis threshold.
  • validated thresholds with sensitivity tests and control experiments.
  • showed that the main results stay consistent across different thresholds.
0
0
4

Delete project?

Are you sure you want to permanently delete this project? This action cannot be undone.

All devlogs, followers, and associated data will be removed.

Followers

Loading…