Astra Suite

AstraDTM & AstraDDG

How a Mutation Changes Stability

Stabilizing a target means finding the few mutations that help among the many that hurt. For any single-point mutation, AstraDDG predicts the change in folding free energy and AstraDTM the change in melting temperature — so the shortlist ranks itself before the bench.

ΔTm (°C)ΔΔG (kcal/mol)Single Mutations
Predicted ΔTm (°C)Measured ΔTm (°C)
190 mutations · 7 receptors·71% correct direction
Predicted vs. Measured ΔTm (Illustrative) · Public GPCR Benchmark Stats Below.
Signature Strength

Beats Structure-Based Predictors — From Sequence

On public ΔΔG benchmarks, AstraDDG ranks first among 27 predictors — outscoring methods that need a solved 3D structure, using sequence alone.

#1
of 27 ΔΔG Predictors
◆ Scored mutation sites

Real Structure · T4 Lysozyme (PDB 2LZM)

Every Mutation, Scored on the Fold

Two Measures of Stability

Free Energy and Melting Point

They answer different questions, and reading both is how you tell a robust change from a lucky one.

AstraDDG · ΔΔG

Thermodynamic stability

The change in folding free energy a mutation causes, in kcal/mol — how much more (or less) the folded state is favoured.

AstraDTM · ΔTm

Thermal stability

The shift in melting temperature a mutation causes, in °C — how much more heat the protein withstands before it unfolds.

On the Structure

See Where Stability Comes From

Stabilizing positions rarely scatter at random — they cluster in the regions that hold a fold together. Scoring every mutation lets those regions surface on the structure itself.

Scored Mutation Sites (Blue) on a 7-TM Receptor Fold

Rank What Stabilizes

The Shortlist, Sorted

Every candidate mutation scored on both axes and ranked — so the handful worth making sits at the top, and the ones that hurt stay off the bench.

MutationΔTmΔΔGVerdict
M2 → I+5.2°C+1.4Stabilizing
L7 → F+3.9°C+1.0Stabilizing
A5 → V+2.1°C+0.6Stabilizing
T9 → S+0.4°C+0.1Neutral
G3 → A-1.8°C-0.7Destabilizing
R4 → E-4.6°C-1.5Destabilizing

Illustrative ranking. Focused on single-point mutations, where the signal is cleanest; combinatorial designs build on the same per-mutation scores.

Benchmarks

Measured Against the Public Field

AstraDDG and AstraDTM are evaluated on the open thermostability benchmarks the field reports on — held-out proteins, no training overlap.

#1 / 27
ΔΔG Predictor Ranking
vs FoldX, ThermoMPNN, RaSP
0.71
ΔΔG Pearson r · GeoStab S461
External Held-Out Benchmark
3.7 °C
ΔTm MAE · Held-Out
75% Correct Direction
ΔΔG Pearson r · public GeoStab benchmarks
GeoStab S4610.71
GeoStab S6690.55

External held-out benchmarks, zero training overlap — and sequence-only, with no experimental structure required.

The sequence-only edge

Structure-based predictors need a solved or modelled 3D structure to score a mutation. AstraDDG works from sequence alone — and still ranks first — so it runs on the many targets that have no structure at all.

GeoStab S669GeoStab S461External Held-out

Across Families

And on the Hardest Membrane Targets

Beyond the single-mutation benchmarks, stability is reported per family in the open Model Performance whitepapers — including ρ 0.93 on T4 lysozyme and directional accuracy across the GPCR superfamily.

190
GPCR Mutations Benchmarked
7 Public Receptors
71%
Correct Direction of Effect
Public GPCR Set
ρ 0.93
T4 Lysozyme Stability
Enzymes Whitepaper

Put Astra on Your Targets

Every Astra model runs inside the Orbion platform. Bring a sequence and get the full read-out — structure, function, modifications, binding, and stability.