In Silico Pathogen Classification Using Vital Signs + Labs — Seeking MIMIC-IV Validation Partners

[netanelcyber]

👋 Introduction

Hi MIMIC community,

I'm the developer of PenuX — an in silico engine for pathogen-specific
classification based on routine clinical parameters (WBC, SpO2, temperature,
HR, RR, CRP, Lactate, Procalcitonin).

Rather than predicting sepsis as a binary outcome, PenuX classifies across
12 specific pathogens (e.g., MRSA, Klebsiella, Pseudomonas, Candida),
providing a ranked differential with an explainable clinical conclusion per
classification — designed as a clinical decision support tool.

---

🎯 What I'm Looking For

I'm seeking collaborators or feedback on:

1. MIMIC-IV cohort extraction — Is there a validated SQL/Python query
   for extracting confirmed bacteremia/fungemia cases with matching vitals
   and lab values from mimiciv_hosp + mimiciv_icu?

2. OMOP CDM alignment — Has anyone mapped MIMIC-IV to OMOP CDM for
   infectious disease cohorts specifically? I saw mimic-omop but it
   appears MIMIC-III focused.

3. Validation partners — Anyone working on pathogen-level (not just
   sepsis-level) ML classification who would be interested in comparing
   cohorts or pipelines?

---

🔬 Current Status

• Engine: Python, XGBoost + rule-based hybrid
• Training data: In silico (synthetic, literature-derived distributions)
• Next step: Real-world validation against MIMIC-IV blood culture results
  (microbiologyevents table)
• OMOP CDM integration: Planned for interoperability

---

📎 Relevant Tables I'm Targeting

Table	Purpose
mimiciv_hosp.microbiologyevents	Ground truth pathogen
mimiciv_hosp.labevents	WBC, CRP, Lactate, PCT
mimiciv_icu.chartevents	SpO2, Temp, HR, RR
mimiciv_hosp.diagnoses_icd	Sepsis/infection ICD codes

---

Any pointers, prior work, or collaboration interest welcome.
Happy to share the PenuX repo for review.

Thanks 🙏