Superspecies AI

Platform Methods

Our workflow integrates computational modeling with experimental microbiology to accelerate antimicrobial discovery. Each stage is designed for reproducibility, interpretability, and alignment with clinical microbiology practice.

1. Data Pipeline & Curation

Data are sourced from public repositories and internal hospital collaborations. Raw files are pre-processed through an automated Snakemake pipeline.

  • Whole-genome assemblies from NCBI RefSeq and PATRIC.
  • Resistance determinants annotated with CARD ontology (β-lactamases, efflux, target mods).
  • Phenotypic MIC values mapped to molecular features for supervised training.
  • Standardized metadata schema using JSON-LD and FAIR principles.
Practical aspect: sequencing data are processed with fastp for trimming and SPAdes for assembly. QC metrics (coverage > 40×, N50 > 50 kb) are enforced before inclusion.

2. Target Identification

Potential targets are ranked by essentiality and druggability scores derived from transcriptomic and structural datasets.

  • RNA-seq differential expression between resistant vs. susceptible isolates.
  • Protein interaction networks modeled using GNN embeddings.
  • Binding-pocket similarity to known druggable sites evaluated via fpocket.
Practical aspect: Essentiality validated by CRISPR-Cas9 knock-outs in E. coli K12; hits showing lethal phenotype at MOI ≥ 0.9 are retained.

3. Generative Design of Molecules & Peptides

We use conditional diffusion and transformer models to generate small molecules and AMPs optimized for multi-objective fitness.

  • Latent conditioning on target pocket embeddings (GraphVAE).
  • Sequence-to-activity model pre-trained on 60 k known AMPs (AMP-DB, DBAASP).
  • Bayesian optimization for balancing potency : toxicity : solubility.
Practical aspect: generated peptides (≤ 25 aa) are synthesized by solid-phase peptide synthesis; purity > 95 % confirmed by HPLC before bioassay.

4. Resistance Evolution Simulation

Molecular-dynamics models simulate probable resistance mutations and efflux up-regulation.

  • In-silico mutagenesis on top 100 binding residues with ΔΔG scoring.
  • Monte-Carlo simulation of population-level mutation fixation under selective pressure.
Practical aspect: validated experimentally via serial-passage evolution of K. pneumoniae clinical isolates under sub-MIC exposure (14 days).

5. ADMET & Toxicity Profiling

Predictive multi-task networks evaluate absorption, metabolism, and toxicity across species.

  • ADMETNet model trained on > 1 M compounds from Tox21 + PubChem BioAssay.
  • Ensemble averaging of physicochemical and neural fingerprint descriptors.
Practical aspect: in-vitro cytotoxicity validated on HepG2 and Caco-2 cell lines; selectivity index ≥ 10 required to progress.

6. In-Vitro & In-Vivo Validation

Lead compounds are evaluated in biosafety-level-2 laboratories following CLSI guidelines.

  • MIC and MBC determined by broth microdilution (CLSI M07-A11).
  • Time-kill kinetics recorded over 24 h using spectrophotometric monitoring.
  • Biofilm inhibition quantified via crystal-violet assay at OD₅₉₀.
  • In-vivo safety assessed in Galleria mellonella larvae prior to murine testing.
Practical aspect: reproducibility ensured through triplicate runs and inclusion of positive (vancomycin) and negative (DMSO) controls.

7. Deployment & Continuous Monitoring

The Superspecies cloud platform continuously retrains models with new genomic data and laboratory feedback.

  • Real-time dashboard built with Supabase and SvelteKit for data versioning.
  • Drift detection module monitors prediction accuracy on external datasets.
  • Federated-learning setup allows secure integration of hospital-level data.
Practical aspect: feedback loops from partner labs trigger weekly incremental retraining and auto-validation against blind test sets.

8. Ethical & Regulatory Compliance

All work complies with biosafety, data-privacy, and ethical-review standards.

  • Institutional Biosafety Committee (IBSC) approval for all microbial experiments.
  • De-identified patient-derived isolates handled under ICMR AMR guidelines.
  • AI models audited for bias and reproducibility under OECD AI Principles.

Our Commitment

Superspecies AI bridges machine intelligence with clinical microbiology. Our goal is to create actionable, safe, and globally accessible antimicrobial innovations.

Collaborate With Us