Wyusyus commited on
Commit
d8e1068
·
verified ·
1 Parent(s): c2955ab

Add Case32 Tdap booster antibody prediction analysis

Browse files

Upload the screened, redacted Open-Science Session and case documentation.

.gitattributes CHANGED
@@ -93,3 +93,4 @@ cases/case25/session.science filter=lfs diff=lfs merge=lfs -text
93
  cases/case28/session.science filter=lfs diff=lfs merge=lfs -text
94
  cases/case27/session.science filter=lfs diff=lfs merge=lfs -text
95
  cases/case29/session.science filter=lfs diff=lfs merge=lfs -text
 
 
93
  cases/case28/session.science filter=lfs diff=lfs merge=lfs -text
94
  cases/case27/session.science filter=lfs diff=lfs merge=lfs -text
95
  cases/case29/session.science filter=lfs diff=lfs merge=lfs -text
96
+ cases/case32/session.science filter=lfs diff=lfs merge=lfs -text
cases/case32/PROMPT.md ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ # Case 32 prompt
2
+
3
+ > How much can baseline immune measurements improve prediction of the antibody response to a Tdap booster beyond basic participant information and pre-existing antibody levels? Use public CMI-PB data to bring together antibody measurements, available immune-cell or plasma-protein features, and participant information. Build a focused prediction analysis for anti-pertussis-toxin response, compare the added value of the different information sources, and explain the biological signals and uncertainties behind the result. Please obtain the public data yourself and leave the sources, analysis and results easy to inspect and rerun.
4
+
5
+ This is the initial research request preserved in the Session export. Subsequent audit and recovery instructions remain in the native Session history rather than being folded into this prompt.
cases/case32/README.md ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Case 32: Baseline immune measurements and Tdap booster antibody response
2
+
3
+ ## Research goal
4
+
5
+ Assess whether baseline immune-cell and plasma-protein measurements improve prediction of the day-14 anti-pertussis-toxin (anti-PT) IgG response to a Tdap booster beyond participant information and pre-existing anti-PT antibody levels. The analysis uses public CMI-PB challenge data and compares models on matched participants with participant-level validation.
6
+
7
+ ## Attribution and package
8
+
9
+ | Field | Value |
10
+ |---|---|
11
+ | Case ID | `case32` |
12
+ | Author / preferred attribution | Wyusyus |
13
+ | Submitter | Wyusyus |
14
+ | Session export date | 2026-10-10 |
15
+ | Status | Draft; independent repository review pending |
16
+ | Model recorded in the Session | `gpt-6-astra` |
17
+ | Package | [`session.science`](session.science) |
18
+
19
+ The native Open-Science package preserves the prompt, Plan, analysis contract, Notebook history, source data, analysis script, report, figures, result tables, review history, and a reproduction bundle. Import it into a compatible Open-Science installation to inspect the saved research; import alone does not rerun the analysis. The original request is also reproduced in [PROMPT.md](PROMPT.md).
20
+
21
+ The uploaded `session.science` is a **separate redacted derivative** of `32-2026-10-10.science`. Eight historical Notebook `PATH` fields contained 176 repetitions of a local Windows user-home prefix. Only those prefixes were replaced with `[REDACTED_USER_HOME]`; the manifest and RO-Crate descriptors were updated accordingly. Scientific inputs, analysis code, results, report, and reviews were not changed. The original export remains local and unchanged.
22
+
23
+ ## Public inputs and rights
24
+
25
+ - [CMI-PB first challenge dataset](https://doi.org/10.5281/zenodo.7702790): baseline antibody, cell, and protein measurements; the Zenodo record identifies its license as CC BY 4.0.
26
+ - [Later CMI-PB dataset and submissions](https://doi.org/10.5281/zenodo.10789473): independent outcome labels; the Zenodo record identifies its license as MIT.
27
+
28
+ The Session report records retrieval on 2026-10-09, source checksums, and the limits of verification of the later archive. Third-party materials retain their own license and attribution requirements; repository inclusion does not change them. This is an analysis of antibody response magnitude, not vaccine efficacy or clinical protection.
29
+
30
+ ## Methods and findings
31
+
32
+ The primary matched multimodal sample contains 18 participants; 57 have eligible baseline/day-14 antibody pairs, and 30 are available for the later-cohort ranking check. Ridge models use repeated, nested participant-level cross-validation with preprocessing confined to training folds. A paired participant-cluster bootstrap describes uncertainty conditional on the fitted cross-validation models.
33
+
34
+ In the matched 18-person comparison, adding cells, proteins, or both to the participant-plus-baseline-antibody reference increased mean held-out RMSE by about 0.098, 0.070, and 0.073 log10 IU/mL, respectively. The paired intervals include modest benefit, so the case does **not** establish equivalence or prove immune features are useless. In the later 30-person cohort, the cell-augmented model ranks outcomes better than the reference point estimate, but the gain is uncertain; baseline antibody alone has the highest reported rank correlation. Cross-assay harmonization limits interpretation of that later-cohort check.
35
+
36
+ The full [Session report](session.science) describes cohort construction, exact metrics, sensitivity analyses, exploratory biological signals, and limitations. In particular, 264 candidate proteins greatly exceed the 18-person multimodal sample. No causal or clinical prediction claim is made.
37
+
38
+ ## Reproduction and verification
39
+
40
+ 1. Import `session.science` into a compatible Open-Science installation and inspect `cmipb-report.md`, `analysis-contract.md`, and `cmipb-analysis-bundle.zip` in the saved artifacts.
41
+ 2. Unpack the analysis bundle in a separate workspace. It includes the public source-table bytes, source manifest, exported `cmipb-analysis.py`, pinned versions, and computed results. Follow its README and run the analysis script from the unpacked directory; no additional data download is required for that rerun.
42
+ 3. Compare the regenerated participant, split, prediction, metric, and uncertainty tables with the saved results. The Session report states that a fresh-namespace rerun in the same managed Python runtime reproduced eight key output tables within tolerance `1e-10`; this is **not** an independently installed-environment test.
43
+
44
+ The original and redacted package inventories were checked against their manifests: all 1,292 inventoried files matched their recorded sizes and SHA-256 hashes. A targeted scan of the redacted outer archive found no local operator name or Windows user-home path. The source-package scan found no private-key header, service-token pattern, or credential-like assignment. Three nested source/bundle ZIPs were readable; the later-release `cmipb-release-v2.zip` is an intentionally selective ZIP-range retrieval rather than a complete standalone ZIP, so it could not be checked as a full archive. One nested upstream source script retains its original `/home/pramod` path; this is from the public source distribution, not the local operator. This screening cannot rule out every sensitive value. Archive checks do not establish scientific correctness.
45
+
46
+ Original export SHA-256: `870770927c30cc93ae13700fef7bdbd8bea32268341fe39dcd2a9bd7d89d0539`.
47
+
48
+ Submitted redacted SHA-256: `84ba12d71955e61006107f57622d740fffe030667ab20a06de48756ff95d7b2f`.
49
+
50
+ Native Open-Science Replay was **not** launched for this export: the recorded recovery found partial file-evidence coverage, and final native readiness remains unconfirmed. A fresh import or full independent scientific rerun of this repository copy is not claimed.
51
+
52
+ ## Sharing and review
53
+
54
+ The submitting user requested publication of this case in the AIPOCH Hugging Face collection. AIPOCH team inspection and public distribution of this submission are within that request, subject to the third-party source terms above. Reuse of case descriptions or figures in a preprint should be confirmed with the author. Independent scientific review and repository acceptance remain pending.
cases/case32/session.science ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:84ba12d71955e61006107f57622d740fffe030667ab20a06de48756ff95d7b2f
3
+ size 6868720630