Aether for drug discovery
In silico — the model proposes
Docking, FEP, ADMET, generative chemistry, pharmacophore, biologics. Reasoning about molecules inside the model.
Aether for drug discovery.
AI that retires the drug-discovery stack.
Aether for drug discovery is what the incumbent computational-chemistry suites collapse into when the model is the substrate. Specialist agents — docking, FEP, MD, QM, ADMET, pharmacophore, generative chemistry, biologics — coordinated by Aether with stable REST and gRPC contracts, so each agent is independently deployable behind your firewall.
- Agents
- 30+
- Path
- Target → Lead
- Modalities
- Small + bio + ASO
- Drives
- Aether for autonomous labs
Coverage
Target to lead, under one model.
The incumbent stack is a dozen seat-licensed tools with a dozen file formats. Aether is one model, one set of contracts and one project file — built so the docking agent feeds the FEP agent feeds the ADMET agent without a single CSV in between.
01
Docking & induced-fit
Pose generation, scoring, induced-fit (IFD) refinement. Ensemble docking against multiple receptor conformations. Calibrated against PDBbind and the Astex Diverse Set.
02
FEP & free-energy
Alchemical free-energy perturbation for relative and absolute binding energies. Replica-exchange with solute tempering (REST). Coupled to MD where convergence demands it.
03
Molecular dynamics
Classical MD with hydration, force-field minimisation, replica exchange. Drives ligand and protein dynamics; runs on established open MD packages or our GPU-native engine.
04
QM / DFT
Quantum chemistry for charges, conformations, transition states, reaction paths. Periodic and embedding methods where the chemistry demands them.
05
ADMET
All major endpoints — solubility, permeability, hERG, CYP inhibition, hepatotoxicity, clearance, half-life, volume of distribution. Calibrated to your assays.
06
pKa & ionisation
Quantitative pKa across functional classes. Drives protomer and tautomer enumeration for downstream docking and FEP.
07
Generative chemistry
Goal-conditioned generation with synthetic-accessibility constraints. Scaffold hopping, lead optimisation, R-group exploration, multi-parameter optimisation.
08
Biologics
Antibody and binder design, paratope prediction, developability scoring (aggregation, viscosity, glycosylation, deamidation), immunogenicity flags.
09
Pharmacophore & 3D-QSAR
Ligand-based pharmacophore generation, hypothesis enumeration, virtual screening, alignment-free 3D-QSAR.
10
Materials & co-crystals
Crystal structure prediction, polymorph search, solubility, co-crystal screening — under the same model as the small-molecule flow.
11
PBPK & translational
Physiologically-based pharmacokinetic modelling, exposure-response, first-in-human dose prediction.
12
ELN & inventory
Electronic notebook with structured assay capture, reagent inventory, plate logistics, hand-off to the autonomous-lab surface for execution.
The loop
Six steps. One model.
From target characterisation through bench readout, the same model handles every step. The loop closes — handing measurements back to the agents that proposed the experiments.
- 01
Target
Pocket characterisation, druggability scoring, hot-spot identification, allosteric site discovery.
- 02
Hit-find
Virtual screening, fragment-based design, generative hits, similarity searches against the catalog.
- 03
Hit-to-lead
FEP-prioritised analogue series, multi-parameter optimisation across potency, selectivity and ADMET.
- 04
Lead-opt
Late-stage SAR, scaffold transitions, ADMET liability fixes, PK/PD modelling, formulation-aware design.
- 05
Bench
Plate plans, reagent binding, instrument allocation — handed off to the autonomous-lab surface for execution.
- 06
Readout
Assay results back into the model. ELN-grade capture, automated curve fitting, IC50/EC50, hit calling.
Modalities
Every modality your portfolio serves.
The same model spans small molecules, peptides, biologics, nucleic acids, degraders and covalent inhibitors. Modality-aware where it matters, shared where it doesn't.
Small molecules
From scaffold to clinical candidate. Generative design, FEP, ADMET, pharmacophore, virtual screening at billion-compound scale.
Peptides & macrocycles
Cyclic peptide design, membrane permeability, oral bioavailability, conformational analysis.
Antibodies & biologics
Paratope design, humanisation, developability, immunogenicity, affinity maturation, ADC linker chemistry.
Nucleic acids
Antisense oligonucleotide and siRNA design, off-target prediction, knockdown efficacy modelling.
Protein degraders
PROTAC and molecular-glue design — ternary complex prediction, linker optimisation, induced proximity scoring.
Covalent inhibitors
Warhead selection, reactivity prediction, selectivity profiling, irreversible-binding kinetics.
Stack it replaces
In place of a dozen.
The typical computational-chemistry stack at a mid-sized biotech today. Aether collapses it. We do not displace every niche tool — but the spine of the workflow now lives behind a single contract.
Commercial docking + scoring
Commercial FEP implementations
Commercial MD packages
Structure-prep tools
Conformer enumeration
Visualisation environments
Workflow-pipelining platforms
Pharmacophore software
Crystallography databases
3D-similarity tools
Open MD scripting stacks
In-house ADMET pipelines
Validation
Numbers against measured reality.
Each row is reproducible from the platform — the structures, the splits, the scoring scripts. We publish the failure modes alongside the wins.
FEP+ benchmark · 8 targets
1.06 kcal
RMSE vs experiment
PDBbind core set · v2020
0.78
Pearson on binding affinity
hERG classification
0.91
ROC-AUC · external set
Solubility (logS) regression
0.45
RMSE · external set
Antibody developability
0.83
ROC-AUC on flag classes
Astex diverse set · docking
78%
<2Å RMSD pose recovery
Vs the comp-chem stack
Target to hit, one model.
The incumbents each own a method — docking, FEP, MD. Aether runs them together and closes the loop with the autonomous lab, so the design and the assay sharpen one model.
✓Generative chemistry
✓Docking & pose prediction
✓FEP / free energy
✓Molecular dynamics
✓ADMET (calibrated to your assays)
✓Biologics / antibody design
✓QM / DFT
✓Closes the loop with the wet-lab
✓One model, target to hit
Why the cycle shortens
Years of discovery, compressed to weeks.
The slowest part of discovery isn't the FEP — it's the handoffs around it. From a target on a Friday to a wet-lab-ready hit list on a Friday three weeks later. Aether collapses the modeling-to-medchem cycle into the autonomous agent flow.
Stage
Today
With Aether
Why
Target to hit list
6–12 months
2–3 weeks
Docking, FEP, ADMET, generative chemistry and re-ranking run as one autonomous study — not a chain of email handoffs between modeling groups.
Hit to lead
9–18 months
8–14 weeks
Multi-parameter optimisation across binding, ADMET, selectivity and synthetic accessibility in a single objective. The optimiser proposes; you decide what enters synthesis.
Lead to candidate
12–24 months
4–8 months
Tighter coupling with the wet lab. Closed-loop active learning, no quarterly hand-offs to medchem.
Authoring CMC / IND modules
weeks
days
Auto-written modules with figures, tables and the model-version hash that produced each calculation — ready for QC review.
Hypothesis to measured hit
9 weeks
Median across the first wave of customer pilots. From target spec to a wet-lab-validated compound on your bench.
Specialist agents
30+
Docking, FEP, MD, QM/DFT, ADMET, pKa, pharmacophore, generative chemistry, biologics — each with stable contracts.
To first clinical candidate
−14 months
Programme-level compression on a typical small-molecule indication. Iteration count per quarter is what compounds, not any one solve being faster.
AI scientists, not chatbots
Aether is an autonomous medicinal chemist.
Not a chat assistant for your computational chemists — a foundation model that does the work. Plans the campaign, picks the methods, runs the agents, validates against your assays, writes the dossier.
- 01
Reads the target
Ingests the receptor, the indication, the prior art and your house chemistry rules. Decomposes the brief into a study tree before any compute runs.
- 02
Picks the right method
Knows when docking is enough and when FEP is required. Knows which water model pairs with which ligand series. The chemist's intuition encoded.
- 03
Runs MD, FEP, ADMET in parallel
Specialist production agents execute the plan as a DAG. Long FEP runs check-point. ADMET predictions ship with calibrated confidence.
- 04
Validates against your data
Cross-checks every prediction against your historical assay endpoints. Drift halts the run and surfaces a citation, not silent miscalibration.
- 05
Writes the dossier
Auto-written candidate dossiers — pose, ΔG, ADMET, synthesis route, IP landscape — with the model-version hash and the run-replay ID.
- 06
Hands off to the autonomous-lab surface
Proposed compounds enter the wet-lab queue as a fully-specified plate map. Active-learning loops feed measurements back into the next round.
Where this leads
A foundation model that discovers, not just chats about discovery — the substrate for general scientific intelligence.
Aether reasons in molecules, proteins, mechanisms and assays — the substrate of biology. We are moving toward general intelligence through the disciplines that have to face physical reality. The model that proposes a compound today is the model that runs the campaign tomorrow.
Closes the loop
The computational half. the autonomous-lab surface is the physical half.
Aether for drug discovery proposes the molecules and the experiments. Aether for autonomous labs binds reagents, schedules plates, drives the instruments, and feeds the measurements back. That's the loop everyone talks about and most platforms still hand off to email.
Aether for autonomous labs
On the bench — the lab confirms
Plate scheduling, liquid handlers + acoustic dispensers + open-source robotics, plate readers, eBR, ALCOA+. Measurements feeding back into the model.
Related research
The work behind this product.
We publish what we learn. The posts below are the substantive notes behind this product — methods, evaluations, case studies.
Start with one programme.
Tell us the target, the modality, and the deadline. We'll come back with a scoped pilot — three to eight weeks, win condition defined together.