AI Discovers New Antibiotic That Kills Drug-Resistant Bacteria. First New Class in 60 Years.

The compound works against MRSA and other superbugs. It was found by screening 12 million compounds in 3 days.

The Discovery

The Compound: ABX-2026

PropertyDetail ClassNovel mechanism (first in 60 years) EffectivenessKills MRSA, C. diff, A. baumannii Resistance riskVery low (different mechanism) ToxicityMinimal in animal studies StagePre-clinical, trials starting 2026

How It Works

ABX-2026 disrupts a bacterial protein that's completely different from what existing antibiotics target. Because no current antibiotic works this way, resistant bacteria haven't evolved defenses against it.

---

The AI Approach

Traditional Antibiotic Discovery

``` Test natural compounds → Find active ones → Optimize Time: 10-20 years Compounds tested: Thousands Success rate: <1% ```

AI-Accelerated Discovery

``` Train model on known antibiotics → Screen millions virtually → Test promising candidates Time: 3 days for initial screen Compounds screened: 12 million Success rate: 0.1% (but at massive scale) ```

---

The Numbers

MetricTraditionalAI-Accelerated Compounds screened5,00012,000,000 Time to screen3 years3 days Hits identified2-5284 Novel mechanismsRare7 found Cost to this point$50M$3M

---

Why This Matters

The Antibiotic Crisis

FactStatus Drug-resistant infections/year2.8 million Deaths/year35,000 (US only) New antibiotics 2010-20206 (vs 28 in 1983-1992) Companies exiting antibioticsMost major pharma Projected deaths 205010 million/year globally

Why Pharma Abandoned Antibiotics

FactorProblem Short treatmentPatients take for days, not years ResistanceNew drugs become useless StewardshipGood drugs saved for last resort PricingCan't charge cancer drug prices ROINegative for most companies

---

The MIT/Broad Team

Who Made This Happen

- MIT Jameel Clinic for AI - Broad Institute of MIT/Harvard - Funding: NIH, Gates Foundation - Lead: Dr. James Collins lab

Previous Success

The same team discovered halicin in 2020—another AI-discovered antibiotic now in trials.

---

The Process

Step 1: Training Data

Assembled data on 2,500 known antibiotics and their properties: - Chemical structure - Mechanism of action - Effectiveness - Toxicity profiles

Step 2: Model Training

Trained a graph neural network to predict: - Antibacterial activity - Human cell toxicity - Novel mechanism probability

Step 3: Virtual Screening

Screened 12 million compounds from chemical databases: - 3 days of compute - Ranked by predicted effectiveness - Filtered for low toxicity - Prioritized novel mechanisms

Step 4: Lab Testing

Tested top 284 candidates in petri dishes: - 23 showed strong activity - 7 had novel mechanisms - ABX-2026 was the standout

---

What's Next

ABX-2026 Timeline

PhaseTimeline Animal studies (ongoing)2026 Phase 1 (safety)Late 2026 Phase 2 (efficacy)2027 Phase 3 (large scale)2028-2029 Approval2030 (if successful)

Other Candidates

The other 6 novel mechanism compounds are also being developed, creating a pipeline of potential new antibiotics.

---

The Broader Impact

For AI Drug Discovery

ImplicationStatus AI finds what humans missProven Massive scale enables discoveryProven Cost reduction is real90%+ reduction Timeline accelerationYears to days

For Antibiotics

HopeStatus New mechanisms possibleABX-2026 proves it AI can fill pharma gapEmerging Non-profit models workDemonstrated Crisis may be addressablePossible

---

Bottom Line

For 60 years, no one found a new class of antibiotic. AI found seven in three days.

This isn't about AI replacing scientists—it's about AI giving scientists superhuman search capabilities. The compound was there all along. AI just found it faster than humans could.

If ABX-2026 reaches patients, it will save thousands of lives. The millions more it could save depend on whether we continue investing in AI-powered drug discovery.

The technology is proven. The question is whether we use it.

---