---
license: mit
language:
- en
pretty_name: ABX-PD-005 Time-Kill Curve Shape Distortion
tags:
- antibiotics
- pharmacodynamics
- resistance
- time-kill
- tabular
task_categories:
- tabular-classification
size_categories:
- n<1k
---

# ABX-PD-005: Time-Kill Curve Shape Distortion

This dataset tests whether you can detect loss of log linear kill coherence.

The curve stops behaving like a clean log linear decline.

It becomes plateaued or multiphasic.

## Time grid

v1 uses

- 0, 2, 4, 6, 24 hours

## Files

- data/train.csv
- data/test.csv
- scorer.py

## Schema

Each row is one timepoint in a time kill series.

Required columns

- row_id
- series_id
- timepoint_h
- organism
- strain_id
- antibiotic_name
- antibiotic_class
- drug_conc_mg_L
- cfu_log10
- media
- assay_method
- source_type
- shape_distortion_signal
- earliest_distortion

## Labels

- shape_distortion_signal
  - 1 for rows at or after distortion

- earliest_distortion
  - 1 only for the first detected row in that series

## Evaluation

Run

- python scorer.py --path data/test.csv

The scorer detects distortion using

- plateau segments
- multiphasic sign changes in slopes
- poor linear fit

The scorer avoids a common mistake.

It does not call distortion when drug concentration changes too much.