Healthcare & Biomedical AI

1. Medical LLMs

General-purpose LLMs perform surprisingly well on medical benchmarks. GPT-4 passed the United States Medical Licensing Examination (USMLE) with a score above 90%. However, medical LLMs fine-tuned on clinical data offer advantages in understanding medical terminology, following clinical reasoning patterns, and generating responses appropriate for healthcare contexts.

2. Clinical NLP Applications

Clinical NLP processes the vast amount of unstructured text in electronic health records (EHRs). Progress notes, discharge summaries, radiology reports, and pathology findings contain critical clinical information that is difficult to query or analyze in text form. LLMs can extract structured data from these notes, identify patients matching clinical trial criteria, detect adverse drug events, and summarize patient histories.

from transformers import pipeline

# Clinical NER using a biomedical model
clinical_ner = pipeline(
    "token-classification",
    model="d4data/biomedical-ner-all",
    aggregation_strategy="simple",
)

clinical_note = """Patient presents with persistent cough and shortness of breath
for 2 weeks. History of Type 2 diabetes managed with metformin 500mg.
Chest X-ray shows bilateral infiltrates. Started on azithromycin
and referred for pulmonary function testing."""

entities = clinical_ner(clinical_note)
for ent in entities:
    print(f"  {ent['entity_group']:>20}: {ent['word']} ({ent['score']:.3f})")

3. Medical Question Answering

from openai import OpenAI

client = OpenAI()

# Medical QA with safety guardrails
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": """You are a medical information assistant for clinicians.
Provide evidence-based answers citing relevant guidelines and studies.
Always note the level of evidence. Flag when a question requires
specialist consultation. Never provide direct patient treatment
recommendations without specifying they need clinical validation."""},
        {"role": "user", "content": """What are the current first-line treatments for
newly diagnosed Type 2 diabetes in adults with HbA1c between 7-8%?"""},
    ],
)

print(response.choices[0].message.content)

4. Drug Discovery and Molecular Generation

LLMs trained on chemical and molecular data can generate novel drug candidates, predict molecular properties, and optimize lead compounds. These models treat molecules as sequences (SMILES notation) and apply the same autoregressive generation techniques used for text. More specialized approaches use graph neural networks or 3D molecular representations, but LLM-based methods benefit from the ability to incorporate textual descriptions of desired properties alongside molecular structures.

# Molecular property prediction with a chemistry LLM
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "seyonec/ChemBERTa-zinc-base-v1"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

# SMILES representation of aspirin
smiles = "CC(=O)Oc1ccccc1C(=O)O"

inputs = tokenizer(smiles, return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)
    prediction = outputs.logits.softmax(dim=-1)
print(f"Molecular properties prediction: {prediction}")

5. Protein Structure and Genomics

Protein language models like ESM-2 (Evolutionary Scale Modeling) treat amino acid sequences as "text" and learn representations that capture protein structure and function. AlphaFold 3 uses a diffusion-based architecture to predict 3D structures of proteins, nucleic acids, and their complexes. These tools are transforming structural biology by enabling rapid structure prediction that previously required months of experimental work.