acceptodds
Under review as a conference paper at ICLR 2027

CalibHyper: Chance-Corrected Relational Hypergraphs for Few-Shot Molecular Property Prediction

Abstract

Molecular property prediction is central to drug development and materials discovery, but experiments are costly and labeled data are scarce. Context-aware methods use auxiliary assay labels to support few-shot prediction. However, label agreement is sensitive to class marginals and does not directly capture dependencies between properties. We propose CalibHyper, a chance-corrected relational hypergraph method based on the joint label distribution. CalibHyper subtracts an independence baseline from the ordered four-state label distribution and applies shrinkage based on the number of joint observations. It estimates residuals through a swap-equivariant relation head, then constructs signed hyperedges conditioned on each molecule, allowing calibrated relation signals to directly inform prediction. Experiments on multiple molecular property prediction datasets show that CalibHyper and its variants achieve competitive predictive performance.

Primary area:
applications to physical sciences (physics, chemistry, biology, etc.)
Keywords:
Molecular Property Prediction, Hypergraph, graph neural network, few-shot learning

Reviews

Among:
Shape:
1234this paper, Top 4%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 3.33 over 3 reviews, above 96% of the others.
Reviewer_mBtYOpenReview ↗
Rating
3(+1)
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes CalibHyper, a chance-corrected relational hypergraph framework for few-shot molecular property prediction. The method introduces C4RC to model four-state property relations while accounting for chance agreement and observation reliability. It further uses MSRHA to select auxiliary properties and propagate signed relation messages for molecular prediction. Experiments on MoleculeNet benchmarks show promising results in both 1-shot and 10-shot settings.

Critical strengths
  1. The paper identifies a meaningful limitation of binary label agreement, particularly under highly imbalanced property labels. The analysis of real assay data provides clear motivation for chance-corrected relation modeling.

  2. The combination of C4RC and MSRHA is well motivated. The proposed framework connects relation calibration with molecule-conditioned auxiliary-property selection and signed message passing, rather than using relation learning only as an auxiliary training objective.

  3. The experimental evaluation covers multiple MoleculeNet benchmarks under both 1-shot and 10-shot settings. The component ablations and additional analyses of auxiliary labels and routing strategies provide useful insights into the proposed framework.

Critical weaknesses
  1. The necessity of four-state relation modeling is not fully established. Since chance correction itself addresses the limitations of binary agreement, a simpler chance-corrected scalar relation could also be effective. Comparing C4RC with such a relation under the same MSRHA framework would better clarify the contribution of four-state modeling.

  2. The reliability shrinkage in Eq. (2) depends only on the number of jointly observed molecules. However, for highly imbalanced datasets such as MUV and PCBA, a large observation count does not necessarily imply reliable estimates of rare label combinations. The authors should clarify whether the proposed shrinkage adequately handles this imbalance.

  3. The effectiveness of C4RC is mainly demonstrated through downstream prediction results. However, the paper provides limited direct evidence that the learned relation scores accurately reflect property dependence. An analysis of the predicted relation signs and strengths, especially for negatively related properties such as those in CEETOX, would better support the motivation of relation calibration.

  4. The empirical benefit of hypergraph-based communication is not fully demonstrated. Although MSRHA jointly encodes molecule-target-auxiliary triples, the current experiments do not directly compare it with simpler relation-aware pairwise message passing. A controlled comparison using the same C4RC relation scores would help clarify the contribution of the proposed hypergraph design.

  5. The effectiveness of C4RC under sparse auxiliary labels remains unclear. In the single-run experiments in Figure 4(a), switching to four-state targets substantially reduces performance on MUV when auxiliary labels are missing. Since reliability shrinkage is intended to account for limited joint observations, a clearer explanation of this behavior would strengthen the analysis.

Questions for authors
  1. In Section 5.2, MSRHA selects auxiliary properties using the largest signed relation scores. How does this strategy retain strongly negative relations when weaker positive relations receive higher scores? Would using the absolute relation scores for Top-k selection be more appropriate?

  2. Table 2 shows that removing MSRHA while retaining C4RC performs worse than removing both modules, especially on MUV. Could the authors explain why relation supervision alone reduces prediction performance and how MSRHA addresses this issue?

  3. In MSRHA, signed hyperedge messages update only molecule and target property nodes, while auxiliary property nodes remain unchanged. What motivates this design, and would updating auxiliary property nodes affect the performance?

  4. In Section 5.1, C4RC uses a support-label-conditioned independence baseline for unseen target properties, which differs from the joint-marginal baseline used for meta-training property pairs. Could the authors clarify how the relation head handles this difference when estimating target-auxiliary relations for query molecules? Does this change in calibration affect the reliability of relation transfer to unseen tasks?

Reviewer_B7WUOpenReview ↗
Rating
3
Confidence
4(+1)
Informativeness
2
Presentation
3
Soundness
3
Read review
Paper summary

CalibHyper targets the problem in few-shot molecular property prediction that “supervising property relations with binary label agreement confounds class marginals with true dependence and merges ordered disagreement states.” It proposes C4RC (chance-corrected four-state relation calibration) and MSRHA (molecule-conditioned signed relational hypergraph adapter). It exceeds the strongest reported baselines in ROC-AUC on 26 dataset/setting conditions, but the gain magnitude is strongly correlated with joint-observation abundance and relational sign structure, with no advantage on NVS where joint observations are sparse.

Critical strengths
  1. Accurate diagnosis with field-level generality. The analysis of binary agreement degenerating under rare positives, so each unit of k changes the target by only 0.0013) is clear and generalizable.
  2. Experimental coverage is above average. Thirteen datasets and 26 conditions in 1-shot/10-shot, with ablations covering chance correction, sign separation, routing count, auxiliary count, and masking.
Critical weaknesses

the contribution lies mainly in the coupling mode, yet the paper lacks controlled experiments on why this coupling must be used rather than a simpler alternative (e.g., would soft weighting by the magnitude of the calibrated residual without sign separation also achieve similar results?)

Questions for authors
  1. Is late-training degradation related to the discontinuity of Top-k selection? Top-k is nondifferentiable where the selected set changes. As training proceeds, the routed sets may flip frequently, increasing the variance of message passing. Can fixing k or freezing the routing sets late in training alleviate the degradation?

  2. Compared with a simpler calibrated weighting without routing scheme, is the additional complexity of MSRHA necessary?

  3. How much of C4RC’s gain comes from chance correction and how much from four-state decomposition?

Reviewer_2ytVOpenReview ↗
Rating
4(+1)
Confidence
1
Informativeness
2
Presentation
2
Soundness
3
Read review
Paper summary

The authors developed CalibHyper, a method for few-shot molecular property prediction that uses labels from auxiliary assays as context. They identify a problem with supervising property relations using binary-label agreement: when almost all molecules are inactive, two unrelated assays agree on nearly every compound, so the agreement reflects the class marginals rather than any real association. Binary agreement also merges the two directional disagreement states. To this end, CalibHyper keeps the four ordered label states, subtracts the agreement expected under independence, shrinks the result when few molecules are jointly measured, and uses the calibrated estimates to select auxiliary properties for each molecule and to set the sign and weight of the messages between them. CalibHyper is evaluated on 13 datasets in 1-shot and 10-shot settings and reports improvements across all conditions.

Critical strengths

The paper lays out the problem and its motivation clearly, and I find the underlying observation about chance agreement in sparse assay data convincing and worth publishing. The evaluation is also broad, covering thirteen datasets in both 1-shot and 10-shot settings against a large set of reported baselines. However, I do not have sufficient background to assess this work, and I recommend assigning it to a reviewer with greater knowledge in this field.

Critical weaknesses

Having no experience in this field, I chose to leave this field empty.

Questions for authors

One thought I have is that the move from binary agreement to four ordered states sounds reasonable and justified. Did the authors consider using the magnitude of each response rather than its direction alone, for instance, the continuous readouts behind the binarized benchmark labels (if such data is available)? As I am not from this field, I do not know what difference it would make, but it seems like a natural next step.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.