Explain why adding full-length CD47 fails to correct reverse docking for D2523

Identify the cause of the instability wherein augmenting the input with the full-length CD47 sequence (including the transmembrane region) fails to normalize docking for the D2523–CD47 complex in AlphaFold 3, despite effectively normalizing the docking direction for the D2510–CD47 complex.

Background

To mitigate reverse docking, the authors supplied AF3 with the full-length CD47 sequence including the transmembrane region. This normalized D2510 docking but produced lower confidence scores and did not resolve the issue for D2523, which still docked incorrectly to an alpha helix.

The authors explicitly note that the reason for this differing behavior across antibodies remains unspecified, leaving open the mechanistic explanation for the instability.

References

However, for the D2523-CD47 complex, the same method was ineffective (D2523 bound to alpha helix), and the reason for this instability has not yet been specified (see Fig. 4).

— Exploring AlphaFold 3 for CD47 Antibody-Antigen Binding Affinity: An Unexpected Discovery of Reverse docking  (2511.14676 - Xu et al., 18 Nov 2025) in Results — Experiment 3 (Reverse Docking Phenomenon)