Highest Single Agent (HSA)¶
The Highest Single Agent model is a simpler alternative to Bliss independence for assessing drug combinations. It asks: does the combination do better than the best single drug alone? The HSA score is the excess over the stronger single agent at the same concentrations.
The Model¶
At each concentration pair \((c_A, c_B)\), the HSA prediction is the response of whichever single drug is stronger at its respective concentration — the response closer to the saturating asymptote \(\text{effect}_\infty\):
- Inhibition (\(\text{effect}_\infty < \text{effect}_0\)): the lower response (more inhibited).
- Activation (\(\text{effect}_\infty > \text{effect}_0\)): the higher response (more activated).
In fractional-effect terms the HSA expectation is \(\max(E_A, E_B)\) in both directions. With single-agent activation of 40 % and 60 %, HSA expects 60 % while Bliss expects 76 %; a combination giving 70 % is synergistic by HSA and antagonistic by Bliss. That is a difference between the two reference models, not a contradiction.
Deviation from HSA¶
where \(\text{effect}_0\) is the zero-drug response and \(\text{effect}_{\infty}\) the saturating response (same asymptotes as Bliss). Same sign convention as Bliss:
| Deviation | Interpretation |
|---|---|
| Negative | Synergy — stronger effect than the better single agent alone |
| Near zero | No excess — combination matches the better single agent |
| Positive | Antagonism — weaker effect than the better single agent |
When to Use HSA vs Bliss¶
- HSA is generally the least stringent benchmark for calling synergy: in fraction-affected terms its expectation \(\max(E_A, E_B)\) is never above the Bliss expectation \(E_A + E_B - E_A E_B\), so a combination beats HSA more easily than Bliss.
- Bliss assumes probabilistic independence and sets a higher bar. A combination can be synergistic by HSA yet antagonistic by Bliss (see the 40 % / 60 % / 70 % example above) — that reflects different null models, not a bug.
Both are computed automatically for matrix analyses.
Further Reading¶
- Wikipedia: Drug combination — overview of null models for combination analysis
- Borisy et al. (2003) "Systematic discovery of multicomponent therapeutics" — high-throughput combination screening in drug discovery (DOI 10.1073/pnas.1337088100; the previous
…2133841100link was a typo)
Implementation details
HSA function (synfit.bliss):
hsa_reference(resp_horizontal, resp_vertical, effect_0=..., effect_inf=...)— element-wise minimum for inhibition, maximum for activation (stronger agent).hsa_deviation(mean_matrix, resp_horizontal, resp_vertical, effect_0, effect_inf)→np.ndarray— normalized deviation from the HSA reference. Same sign convention asbliss_deviation(): negative = synergy, positive = antagonism.resp_horizontalis a 1-D array of shape(n_hor,)— the horizontal drug's response at each concentration (taken from the first row of the mean matrix).resp_verticalis a 1-D array of shape(n_ver,)— the vertical drug's response at each concentration (taken from the first column of the mean matrix).