CORAL-AI PANGEA reduced 500,000 compounds and found a combination with an MIC of 8.03 µg/mL

Executive summary

From 500,000 compounds to a prioritized combination

ISR-BIO used CORAL-AI PANGEA to define a target—kept anonymous in this case—and screen 500,000 natural compounds. In two weeks, it prioritized 14 candidates and proposed studying the combination of Candidates A and B. Validation against isolate PS011 confirmed bactericidal and bacteriostatic candidates: Combination A + B reached an MIC of 8.03 µg/mL, while a 4PL fit placed it as the lowest-concentration response (IC50 2.19 µg/mL; IC80 4.16 µg/mL), with R² above 0.99 for all five fitted profiles.

Modality

Virtual screening with CORAL-AI PANGEA plus in vitro validation

Library

500,000 natural compounds

Validation model

P. salmonis PS011, the causative agent of piscirickettsiosis

Primary metric

MIC at 80% inhibition and MBC by subculture

Configuration

1:2 serial dilutions, five replicates, seven days, OD620 readout

Key result

14 prioritized candidates; Combination A + B lowered IC50 by 36.8% versus the individual candidate

01 · Context

Why screen half a million compounds before entering the lab

The search did not begin on a culture plate. CORAL-AI PANGEA covered a chemical diversity that would have been impractical to test entirely in the laboratory and prioritized molecules most likely to inhibit growth while also fitting as precision functional additives. Two weeks later, the search space had been reduced to 0.0028% of the original library.

Experimental validation separated two different questions. MIC asks whether a candidate inhibits at least 80% of growth; MBC, assessed by subculture after treatment removal, distinguishes a bacteriostatic response from bactericidal action. Four-parameter logistic (4PL) dose–response curves generated interpolated IC50 and IC80 values that complement—but do not replace—observed MICs.

02 · Challenge

Comparing candidates with different experimental windows

Candidates had maximum solubilities between 45 and 500 µg/mL in IFOP-PsM11. Therefore, the absence of an MIC could mean lack of activity, but it could also mean that the candidate did not reach a sufficiently high concentration in aqueous medium.

The combination showed precipitate above 129 µg/mL, and those points were excluded from analysis. The challenge was to identify reproducible signals without extrapolating beyond the measured range.

03 · Approach

From observed inhibition to dose–response profiles

The PS011 isolate was cultured to exponential phase, adjusted to 0.5 McFarland, and exposed for seven days to 1:2 serial dilutions in 96-well plates at 18 °C and 120 rpm. Growth was quantified by OD620 with five replicates per condition; differences from control were analyzed with one-way ANOVA at p < 0.05.

Cultures without apparent growth were subcultured with 3 µL on AUSTRAL TS-Hem agar to establish MBC. Five candidate curves were then fitted with a 4PL model, excluding precipitated points.

04 · Results

The combination concentrated the strongest inhibitory signal

In the first campaign, two candidates reached MICs of 62.5 and 31.25 µg/mL, respectively, and both showed an MBC of 125 µg/mL. In the second campaign, a third candidate showed a bactericidal profile, with an MIC of 62.5 µg/mL and an MBC of 125 µg/mL. Two others were bacteriostatic, with MICs of 10.4 and 500 µg/mL.

The 4PL fit placed Combination A + B as the lowest-concentration response, followed by Candidate A. No interpolation replaces an observed MIC, but the direction was consistent: the combination ranked first in MIC, IC50, and IC80, and all fits reached R² > 0.99 on observed growth.

IC50 and IC80 values are in µg/mL and interpolated with a 4PL model. They must not be read as observed MIC or MBC values, or as evidence of in vivo activity.

05 · Notable candidates

The functional profiles that ordered the panel

Screening separated the 14 candidates into three groups: candidates without an MIC within the soluble range, bacteriostatic candidates, and bactericidal candidates confirmed by subculture. The first group does not mean inactivity: in several cases, maximum solubility prevented conclusive concentrations from being reached.

Among profiles with a signal, Candidate A was the most potent individual candidate in the 4PL fit (IC50 3.46 µg/mL), while Combination A + B shifted both thresholds to lower concentrations with a robust fit (R² = 0.997). The confirmed bactericidal profile—MIC 62.5 µg/mL and MBC 125 µg/mL—shows that the panel contains both lethal-action and containment candidates, two distinct experimental utilities.

06 · What it means

A broad panel reduced to candidates that can be prioritized

The screen supports prioritization by functional profile, not only by a single optical-density decrease: potency (IC50/IC80), solubility window, and action type (bactericidal versus bacteriostatic) are separate, measurable criteria.

The synergy signal from Combination A + B justifies studying it at higher resolution. Next steps are to confirm it with independent biological replicates, formally quantify the A–B interaction, and assess safety in salmon cells. These data show in vitro activity; they do not establish therapeutic efficacy or a dose for fish.

07 · Limitations

Experimental scope

The assays used a single isolate (PS011); activity against other P. salmonis isolates has not been established. Solubility windows differed among candidates (45–500 µg/mL), so absence of an MIC cannot distinguish inactivity from insufficient exposure. The combination precipitated above 129 µg/mL and was not extrapolated beyond the measured range.

IC50 and IC80 are 4PL interpolations, not observed MIC or MBC values. Synergy is a compatible signal, not a formal demonstration: it requires measuring both components separately under a combination design such as Loewe, Bliss, or FIC. The demonstrated activity is in vitro; in vivo efficacy and fish dosing are outside this study’s scope.

Sources: bacteriology reports No. 08/2026 and No. 009/2026; data and analytical outputs from ISRBIO_01_SRS-DATA.