Quantum Cognition Machine Learning for Forecasting Chromosomal Instability
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.
Ines: We're looking at the paper: "Quantum Cognition Machine Learning for Forecasting Chromosomal Instability".
Title and authors: Ines: So we're looking at this paper now called "Quantum Cognition Machine Learning for Forecasting Chromosomal Instability." It sounds like they're using some really advanced quantum concepts to predict if a circulating tumor cell, or CTC, has high metastatic potential just by looking at its shape.
Marcus: Right. The authors are Di Caro and Kirakosyan and Abanov and Busemeyer and Candelori and Hartmann Lam Musaelian Samson Steinacker Villani Wells Wenstrup Xu. They're bringing in this quantum machine learning framework to tackle a problem that's really hard in liquid biopsy diagnostics, which is detecting these rare cells among millions of white blood cells.
Yuki: I wonder what kind of biological information they think is being captured by just the morphology, because if it works on just shape, does it actually reflect the underlying genomic instability?
Ines: That's the core question. The paper sets up this whole system where they treat data points as quantum states in a complex Hilbert space and use an error Hamiltonian to find a state vector that represents the data point, which is what they call psi t>
Marcus: And they map the features of that CTC—like nuclear characteristics and intensity measures—to these quantum observables, which are Hermitian operators, and then they optimize this objective function to learn those observables.
Yuki: So it's like the machine is trying to find a quantum description of what instability looks like in a cell, instead of just looking at pixels in a standard way?
Ines: Exactly. The paper says that QCML models create this global quantum manifold model of the original data manifold, which means it should be robust to noise and generalize well beyond the training samples they use >
Marcus: They also introduce something new called the QCML Positive Operator-Valued Measure, or POVM, which is an extension that lets them forecast a probability density function for the target variable instead of just a single point estimate >
Yuki: A probability density function sounds useful because it gives you more than just a yes or no prediction; it gives you confidence intervals on how likely that instability level is to be high or low.
Ines: And they use this POVM to forecast both real-valued predictions for the large scale state transition, which is called pLST, and probability forecasts for whether the pLST status is above or below a certain threshold >
Marcus: The performance comparison shows that QCML outperforms some other machine learning methods when predicting pLST CTC in out of sample verification, achieving the highest balanced accuracy and specificity compared to a model like SVM with Gaussian Kernel >
Yuki: That's interesting because it suggests this quantum approach is more reliable than classical models in this high-dimensional, low-sample-size context that liquid biopsy often operates in >
Ines: The paper also digs into feature importance using a gradient-based approach, showing what features the model actually picked up as important for predicting the level of chromosomal instability >
Title and authors: Marcus: They found that features like Cross channel Local Binary Pattern, which is comparing CK against DAPI, and cellular colocalization between CK and DAPI were predictors of low chromosomal instability >
Yuki: So it's not just one feature; it's how different channels interact with each other spatially within the cell that seems to matter for predicting instability >
Ines: They also highlighted the signal from lower DAPI quartile intensity from the cellular mask, which they said indicates the cytoplasmic signal is a really important feature in predicting higher levels of chromosomal instability, or LST+ >
Marcus: And they used a QCML distance metric to define a similarity measure between two data points based on quantum fidelity, and this distance was much better at finding the separation between genomically unstable and stable CTC than the standard Euclidean distance >
Yuki: So what does that actually mean for someone listening who just wants to know if this is useful in practice? Does it mean we can start screening patients sooner with better accuracy?
Ines: It suggests that CTCs with chromosomal instability are likely larger in cellular size and show higher spatial discordance between their nuclear and cytoplasmic structures >
Marcus: They abstract a model for these instability-driven morphological shifts where the cells are bigger and have more misalignment between the nucleus, the cytoplasm, and the cytoskeleton >
Yuki: That points toward a biological mechanism where poor alignment of these structures is what drives that chromosomal instability we're trying to detect >
Ines: The paper also suggests that this cellular localization of DAPI intensity and size is especially important for predicting chromosomal instability in general >
Marcus: So, the main conclusion here is that this QCML-based pLST model outperformed conventional machine learning approaches in predictive accuracy, which is a big claim given the complexity of the method >
Yuki: It's still early days, though. They state that future studies will be required to validate these findings and establish whether using this QCML-based pLST model can actually predict patient survival with better performance compared to conventional methods >
Ines: That sounds like the next big step, moving from prediction of instability to predicting actual patient outcomes >
Marcus: It's a lot of work, but the fact that they managed to build this quantum-like framework and show it has superior out of sample performance is certainly worth paying attention to in terms of what machine learning can achieve >
Yuki: So we'll keep an eye on those validation studies because if this holds up, it could really change how we use liquid biopsies for high metastatic potential cells.
Ines: We will definitely be watching that next phase closely, because the QCML framework itself shows promise in handling these very complex, noisy biomedical data sets >
The paper's summary: Ines: So, to recap, this paper takes these complex images of tumor cells and uses quantum mechanics—specifically Quantum Cognition Machine Learning—to predict if that cell is going to have a high level of chromosomal instability.
Marcus: Exactly. They’re basically treating the biological image data like a quantum state in a Hilbert space, which is way more abstract than just using standard machine learning algorithms on pixel counts.
Yuki: What this actually means for us in the lab is that we don't have to hand-pick every single feature we think matters; the QCML framework helps it find patterns that are biologically relevant but hidden in the noise.
Ines: Right. It’s about finding a way to model these instability shifts where cells get bigger and their internal structures, like the nucleus and the cytoplasm, become poorly aligned >
Marcus: And they use this distance measure between data points, which is based on quantum fidelity, to actually separate the stable cells from the unstable ones better than just using standard Euclidean distance.
Yuki: That separation is key because if you can reliably distinguish those two groups using a model that understands spatial discordance, it supports the idea that poor physical organization drives genomic instability >
Ines: They also found that features related to how different channels overlap—like comparing the signal from CK against DAPI—are actually important predictors for low instability >
Marcus: And they even flagged the intensity of the DAPI signal in the cytoplasmic mask as a key feature, suggesting that what's happening in the cytoplasm is just as telling as what's inside the nucleus >
Yuki: It’s interesting because this shifts our focus from just looking at one structure to understanding how all these different cellular components interact spatially to determine instability >
Ines: The real payoff here is the POVM they use, which lets them forecast a full probability density function for that instability level, not just a single guess >
Marcus: That means they can give us confidence intervals on their prediction, telling us exactly how sure the model is about whether a cell will be unstable or stable >
Yuki: So if this method proves robust in these high-dimensional, low-sample-size contexts you mentioned earlier, it opens up a new way to screen for aggressive cancers in liquid biopsies >
Ines: It certainly shows that this quantum modeling approach can handle the complexity of CTC morphology without requiring us to manually curate every single input variable >
Marcus: But we have to remember the caveat they mentioned: this is still preliminary, and they need more validation before we can claim it’s perfectly accurate for clinical use >
Yuki: True. The next step is seeing if this predictive power translates into patient survival data, which is what they say in their conclusion >
Ines: That's what we're looking toward next, moving beyond just predicting instability to actually predicting patient outcomes with this new tool.
The paper's improvements: Tom: So, we’ve heard that Ines and Marcus talk about how this paper uses quantum methods to predict chromosomal instability in circulating tumor cells, and now they’re going to get into what improvements the authors suggest for this QCML model itself.
Ines: Right. The authors aren't just happy with the initial results; they are proposing ways to make the whole framework more robust and useful >
Marcus: They focus on a few things here: first, they introduce something called the Quantum Cognition Machine Learning Positive Operator-Valued Measure, or POVM >
Yuki: And that POVM is big because it moves beyond just guessing a single instability score; it allows them to forecast a whole probability density function for the instability level >
Ines: That means they can give us confidence intervals, telling us exactly how sure the model is about whether a cell will be above or below a certain threshold >
Marcus: They also mention using this POVM with a mixed loss function that blends L1 and cross-entropy terms, which helps balance out the sensitivity and specificity of their classification >
Yuki: That’s smart. It shows they aren't just optimizing for one thing, but trying to get a good middle ground between being right when it matters most >
Ines: And then there’s the feature importance part, which they refine by using a gradient-based approach to rank the morphology features that matter most for predicting instability >
Marcus: They specifically highlight things like Cross channel Local Binary Pattern for comparing CK against DAPI and cellular colocalization between CK and DAPI as being strong predictors of low instability >
Yuki: That links back to what we discussed—it’s not just the shape of the nucleus, but how it interacts spatially with the cytoplasm that matters for predicting genomic shifts >
Ines: They also found that the QCML distance metric is superior because it finds a better separation between unstable and stable cells compared to a standard Euclidean distance >
Marcus: So these improvements suggest they're trying to make the model not only more accurate but also more interpretable and reliable across different data points in their cohort >
Yuki: It sounds like they’re moving toward a tool that isn't just a predictor, but one that understands the underlying physical rules governing how cells become unstable >
Ines: Exactly. They are building this model to be better at finding those subtle morphological shifts we talked about earlier, and hopefully, it will actually help us distinguish between healthy and highly aggressive tumors in a liquid biopsy setting >
Conclusion: Ines: So we’ve covered how this paper uses quantum mechanics to predict chromosomal instability in circulating tumor cells, and now we’re wrapping up with what it all means for real-world diagnostics >
Marcus: It really shows that using these quantum modeling techniques can give us a predictive edge over standard machine learning methods when the sample size is small, which is exactly what we see in liquid biopsies >
Yuki: I just think the implication for population genetics is that if we can reliably flag cells with high chromosomal instability this early, it could change how we monitor cancer progression across different patient groups >
Ines: Right. The main thing to remember about this paper, "Quantum Cognition Machine Learning for Forecasting Chromosomal Instability," is that it’s not just a theoretical exercise; they showed superior out of sample verification performance compared to SVM and other models >
Marcus: Yeah, the numbers they put out for balanced accuracy and specificity are quite strong when you compare it to classical approaches, especially in those high-dimensional contexts >
Yuki: It supports the idea that biological complexity often requires a more sophisticated mathematical representation than just simple linear feature mapping to capture those subtle spatial relationships between nuclear and cytoplasmic structures >
Ines: Exactly. They found that features like protein correlation between channels and cellular colocalization are actually the most important predictors for low instability, which is what we expected biologically >
Marcus: But the real kicker is how they used the QCML distance metric; it’s a better measure of similarity between two data points than just Euclidean distance >
Yuki: That fidelity aspect suggests that these quantum models are actually capturing some kind of underlying physical coherence in how those cells are structured, not just surface-level pixel differences >
Ines: It’s a powerful tool for sequence and structure analysis because it lets us look at morphology as a quantum system rather than just a set of static measurements >
Marcus: So while they say this is preliminary, the fact that they developed this framework with its POVM forecasting capability is something we need to keep an eye on >
Yuki: And I think the next big step for this research will be those validation studies where they check if predicting chromosomal instability actually translates into better patient survival metrics >
Ines: That’s the goal, right? Moving from just predicting a state transition to making a direct link to clinical outcomes >
Qognitive, Inc. · Epic Sciences · Stony Brook University · Wayne State University · Cornell University · New Jersey Institute of Technology · King’s College London · §Indiana University · §University of Vienna
q-bio.QM, cs.LG, quant-ph
Submitted: 2025-06-02
Updated: 2026-11-16
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
Importance score: 80/100
The gist: The accurate prediction of chromosomal instability from the morphology of circulating tumor cells (CTCs) enables real-time detection of CTCs with high metastatic potential in the context of liquid…
Key concepts
- QCML
- A machine learning approach rooted in quantum theory where data points are treated as quantum states in a complex Hilbert space. This framework allows for context-aware modeling and enhanced generalization without needing extensive manual feature selection.
- POVM
- Positive Operator-Valued Measure, an extension of QCML that enables the system to forecast probability density functions instead of single predictions. This provides confidence intervals and allows forecasting whether instability will exceed a specific threshold.
- QCML Distance
- A similarity measure defined by the quantum fidelity between two data points. This metric proved superior to standard Euclidean distance in effectively separating genomically unstable CTCs from stable ones, helping the model find better class separation.
- Feature Importance
- A method used to rank which specific morphological and protein features (like cross-channel LBP or DAPI intensity) most influence the final prediction of chromosomal instability. This helps researchers understand the biological mechanisms driving cell instability.
Terminology
Summary
The accurate prediction of chromosomal instability from the morphology of circulating tumor cells (CTCs) enables real-time detection of CTCs with high metastatic potential in the context of liquid biopsy diagnostics This preliminary findings support the application of QCML as a novel machine learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts. p. 0 These preliminary findings support the application of QCML as a novel machine
learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts. QCML
enables the simulation of cognition-like learning for the identification of biologically meaningful prediction of
chromosomal instability from CTC morphology, offering a novel tool for CTC classification in liquid biopsy.Where the paper says it
Data Acquisition and Feature Extraction
The study utilizes a pipeline to process high-resolution fluorescence images acquired via the ZEISSTM Axio automated scanning system to identify rare candidates for CTCs among millions of white blood cells A multiscale feature enhancement algorithm helps identify nuclei while cell segmentation occurs across various fluorescence channels (CK, DAPI, CD45/CD31) A total of 112 morphology features were extracted as previously published These features include eight characteristics from each of the nuclear and cell masks and 44 intensity features from nuclear and cell masks across the three DAPI, CK and CD45/CD31 channels. Furthermore, 70 texture features are extracted using Gabor filters to characterize image patterns. p. 1 70 texture features are extracted to characterize image patterns. Gabor filters were extracted
for localized frequency and orientation information in images and to detect irregularities in repeating textures with a fractal feature
approximation.Where the paper says it
Quantum Cognition Machine Learning Framework
QCML represents data observations as quantum states in complex Hilbert space The error Hamiltonian is defined as H(xt) = 1/2 Sum k (Ak - xt,k · I) squared Data points are mapped to quantum states ψt⟩ by finding the ground state of the error Hamiltonian In the supervised setting, the target variable y ∈ R is assigned an N-dimensional quantum ‘forecast’ observable B The corresponding forecast is given by ybt = ⟨ψtBψt⟩.
Model Performance Comparison
QCML outperforms SVM and other machine learning methods when predicting pLST CTC in out of sample verification which achieves the highest balanced accuracy and specificity compared to SVM with Gaussian Kernel Among classification models, QCML POVM achieves the highest AUC-ROC score. In regression tasks, QCML shows a higher out-of-sample specificity (57%) concordance to the ground truth while achieving a high sensitivity of 84% and outperforms the rest of the models in terms of balanced accuracy (70%). p. 6 Showing the average and standard deviation of sensitivity,
specificity and balanced accuracy across 25 folds (5-fold with 5 repeats). QCML shows the highest out-of-sample specificity (57%) concordance to the ground truth while achieving a high sensitivity
of 84% and outperforms the rest of the models in terms of balanced accuracy (70%).Where the paper says it
Feature Importance Insights
QCML identified that Cross-channel LBP and Cell colocalization features were predictors of low chromosomal instability, which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic structures Interestingly, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+).
Visualization and Interpretation
The QCML distance between two data points xt, xt' is defined as dQ(xt, xt') = 1 − ⟨ψtψt'⟩ squared Using this distance matrix in multidimensional scaling (MDS) visualization shows that the QCML distance is much better at finding a separation between genomically unstable and stable CTCs. The model learned that CTCs are larger in cellular size with higher spatial discordance between nuclear and cytoplasmic structures This suggests that CTCs with chromosomal instability may be more likely to have poorly aligned nuclear, subcellular and cytoskeletal textures. p. 8 As shown in Figure 6, the top
10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of:
(i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and CK vs CD45/CD31
and cellular colocalization between CK-DAPI;
(ii) Intensity features of the DAPI cellular mask;
(iii) Fractal nuclear and cellular features. 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 2) Interestingly, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the
cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+).Where the paper says it
Where the paper says it
Conclusion and Future Directions
The QCML-based pLST model outperformed conventional machine learning approaches in predictive accuracy The ultimate objective is to enable real-time detection of CTCs with elevated level of chromosomal instability and high metastatic potential. Future research should explore quantum-assisted feature mapping to identify complex and subtle patterns directly from raw image pixel data. QCML learned a gradient of feature importance that was found posthoc to be biologically and mechanistically involved with chromosomal instability in cancer cells. p. 0 QCML
enables the simulation of cognition-like learning for the identification of biologically meaningful prediction of
chromosomal instability from CTC morphology, offering a novel tool for CTC classification in liquid biopsy. Keywords: Quantum Cognition Machine Learning (QCML), Machine learning, Liquid biopsy, Chromosomal instability,
Circulating tumor cells (CTCs), Metastatic breast cancer
1 Background
Unlike traditional tissue tests[1, 2], cell-based liquid biopsy assays enable selection of individual CTCs for the analysis of chromosomal instability using next-generation sequencing by quantification of large-scale state transitions (LST) [3–9]. 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 2) Interestingly, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the
cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+). As shown in Figure 6, the top
10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of:
(i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and CK vs CD45/CD31
and cellular colocalization between CK-DAPI;
(ii) Intensity features of the DAPI cellular mask;
(iii) Fractal nuclear and cellular features. 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. As shown in Figure 6, the top
10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of:
(i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and CK vs CD45/CD31
and cellular colocalization between CK-DAPI;
(ii) Intensity features of the DAPI cellular mask;
(iii) Fractal nuclear and cellular features. 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 2) Interestingly, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the
cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+). QCML
outperforms conventional machine learning methods when tested on out of sample verification CTCs, achieving higher accuracy in identifying predicted large-scale state transitions (pLST) status from CTC-derived
morphology features. These preliminary findings support the application of QCML as a novel machine
learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 1 ---
Quantum Cognition Machine Learning for Forecasting Chromosomal Instability The accurate prediction of chromosomal instability from the morphology of circulating tumor cells (CTCs) enables real-time detection of CTCs with high metastatic potential in the context of liquid biopsy diagnostics. QCML leverages quantum mechanical principles to represent data as state vectors in a Hilbert space, enabling context-aware feature modeling, dimensionality reduction, and enhanced generalization without requiring curated feature selection. p. 0 Here, we introduce the application of Quantum Cognition Machine Learning (QCML), a quantum-inspired computational framework, to estimate morphology-predicted chromosomal instability in CTCs from patients with metastatic breast cancer. QCML leverages quantum mechanical
principles to represent data as state vectors in a Hilbert space, enabling context-aware feature modeling,
dimensionality reduction, and enhanced generalization without requiring curated feature selection. QCML leverages quantum mechanical
principles to represent data as state vectors in a Hilbert space, enabling context-aware feature modeling,
dimensionality reduction, and enhanced generalization without requiring curated feature selection. QCML
outperforms conventional machine learning methods when tested on out of sample verification CTCs, achieving higher accuracy in identifying predicted large-scale state transitions (pLST) status from CTC-derived
morphology features. QCML
outperforms conventional machine learning methods when tested on out of sample verification CTCs, achieving higher accuracy in identifying predicted large-scale state transitions (pLST) status from CTC-derived
morphology features. These preliminary findings support the application of QCML as a novel machine
learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts. These preliminary findings support the application of QCML as a novel machine
learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts. QCML
enables the simulation of cognition-like learning for the identification of biologically meaningful prediction of
chromosomal instability from CTC morphology, offering a novel tool for CTC classification in liquid biopsy.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 2 ---
Quantum Cognition Machine Learning (QCML) [25–28] is an emerging field that introduces a novel approach to machine learning, grounded in the mathematical principles of quantum theory In QCML, data points are represented as quantum states in a complex Hilbert space The inner product between two states ψ1⟩ and ψ2⟩ is denoted by the bra-ket ⟨ψ1ψ2⟩. A measurement of a quantum observable, represented by a Hermitian operator M, in the state ψ⟩ yields an eigenvalue mi of M with probability given by the squared magnitude of the overlap with the corresponding eigenstate mi⟩: ⟨miψ⟩ 2. p. 1 The inner product between two states |ψ1⟩ and |ψ2⟩ is denoted by the bra-ket
⟨ψ1|ψ2⟩. A measurement of a quantum observable, represented by a Hermitian operator M, in the state |ψ⟩ yields an eigenvalue
2 The
expression ⟨ψ|M|ψ⟩ gives the expected value of the random variable associated with measuring M in the state |ψ⟩ [35, p. In QCML, for each vector xt ∈ RK belonging to a data set consisting of t = 1, . In QCML, for each vector xt ∈ RK belonging to a data set consisting of t = 1, . , T observations, we define an error Hamiltonian
as
H(xt) = 1
2
X
k
(Ak − xt,k · I)
2
. Each of these K quantum
observables can be viewed as a ‘quantization’ of a corresponding feature of the original K-dimensional data set. The vector xt
then can be mapped to a quantum state |ψt⟩ by finding the ground state (i. The vector xt
then can be mapped to a quantum state |ψt⟩ by finding the ground state (i. , the eigenstate associated with the lowest eigenvalue)
of the error Hamiltonian (1).Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 3 ---
In unsupervised setting, training a QCML model involves iterative updates to the observables Ak so that the ground states ψt⟩ ‘cohere’ to the data, that is, the distance between xt and its position x(ψt) is minimized, as well as the variance of the measurement. In a supervised setting, which is the main focus of this article, training process differs from unsupervised case The target variable y ∈ R is assigned an N-dimensional quantum ‘forecast’ observable B. p. 2 In the supervised setting, which is the main focus of this article, the training process differs from unsupervised case [28]. The
target variable y ∈ R is assigned a N-dimensional quantum ‘forecast’ observable B. The
target variable y ∈ R is assigned a N-dimensional quantum ‘forecast’ observable B. Given a data point xt the corresponding
forecast, measured in quantum state ψt, is given by
ybt = ⟨ψt|B|ψt⟩ .Where the paper says it
Where the paper says it
--- Page 4 ---
The QCML Positive Operator-Valued Measure (POVM) extends QCML to predict probability density functions for targets instead of single point estimates This extension provides the ability to forecast full probability distributions, which enables the estimation of confidence intervals and offers a more detailed understanding of the predictions. QCML POVM allows us to forecast the probability density function p(y) of a continuous target variable y instead of point estimates. p. 3 For a state |ψ⟩, the probability of observing the outcome associated with
the POVM element k is given by pk = ⟨ψ| Fˆ
k |ψ⟩. QCML POVM allows us to forecast the probability density function p(y) of a continuous target variable y instead of point
estimates.Where the paper says it
--- Page 5 ---
QCML POVM allows us to forecast both real-valued forecasts of pLST (based on expected mean/median) and probability forecasts of pLST being above/below a certain threshold The mixed-loss function takes the following form: LTotal = LL1 L(gradient-free) L1 + LCE L(gradient-free) CE, (2).
--- Page 6 ---
QCML POVM is an extension of QCML that allows one to forecast probability density functions for targets instead of single point estimates The set of operators Fˆ(y) = Yˆ †(y)Yˆ (y), indexed by the continuous parameter y, forms a POVM. By construction, the probability density function p(y) is given by p(y) = ⟨ψ Yˆ †(y)Yˆ (y) ψ⟩. p. 3 ˆ (3)
The set of operators Fˆ(y) = Yˆ †
(y)Yˆ (y), indexed by the continuous parameter y, forms a POVM. By construction, the probability
density function p(y) is given by
p(y) = ⟨ψ| Yˆ †
(y)Yˆ (y) |ψ⟩ .Where the paper says it
--- Page 7 ---
Figure 4: ROC AUC curves for the top three performing models: QCML POVM (blue), XGBoost (red), and Random Forest (green) QCML POVM achieves the highest ROC AUC score of (0. p. 1 QCML outperforms SVM and other
machine learning methods when predicting pLST CTC in out of sample verification which achieves the highest balanced accuracy
and specificity compared to SVM with Gaussian Kernel, the best performing among classical models. Additionally, among the
classification models, QCML POVM achieves the highest AUC-ROC score, a threshold-independent performance metric. QCML POVM achieves the highest ROC AUC score of (0. 763) as shown in Table 2 and Figure 4 with ROC AUC curves for top
performing models. QCML POVM achieves the highest ROC AUC score of (0. 763) as shown in Table 2 and Figure 4 with ROC AUC curves for top
performing models. QCML POVM achieves the highest ROC AUC score of (0. 763) as shown in Table 2 and Figure 4 with ROC AUC curves for top
performing models.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 8 ---
Figure 5: QCML POVM produces full probability distribution of LST for all observations Having probability distribution allows one to forecast both LST quantiles as well as probabilities for LST being above/below a threshold.
--- Page 9 ---
QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI. Additionally, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+). p. 8 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 2) Interestingly, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the
cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+).Where the paper says it
--- Page 10 ---
Figure 6: QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI.
--- Page 11 ---
QCML distance is a type of supervised similarity measure, since the representation of the data in quantum states ψt has been optimized using the training targets The QCML distance is much better at finding a separation between genomically unstable and stable CTCs. The texture cross-channel measures of colocalization identified by QCML suggest that CTCs with chromosomal instability may be more likely to have poorly aligned nuclear, subcellular and cytoskeletal textures.
--- Page 12 ---
QCML’s prediction of chromosomal instability classified by QCML abstracted a model for instability-driven morphological shifts where CTCs are larger in cellular size with higher spatial discordance between nuclear and cytoplasmic structures. QCML findings were also corroborated by the evidence that the cellular localization of DAPI intensity and size is important for the prediction of chromosomal instability. p. 8 1) Cross-channel LBP and Cell colocalization features in the QCML model were predictors of low chromosomal instability
which is in line with those features’ ability to measure cellular remodeling and spatial discordance between nuclear and cytoplasmic
structures [42]. 2) Interestingly, QCML weighed the signal of the lower DAPI quartile intensity from the cellular mask which indicates the
cytoplasmic signal as one of the most important features in predicting higher levels of chromosomal instability (LST+).Where the paper says it
--- Page 13 ---
Future studies will be required to validate these findings and establish whether the presence of CTC with predicted chromosomal instability classified by QCML can predict patient survival with better performance compared to conventional methods. The QCML-based pLST model outperformed conventional machine learning approaches in predictive accuracy. p. 0 QCML leverages quantum mechanical
principles to represent data as state vectors in a Hilbert space, enabling context-aware feature modeling,
dimensionality reduction, and enhanced generalization without requiring curated feature selection. QCML
outperforms conventional machine learning methods when tested on out of sample verification CTCs, achieving higher accuracy in identifying predicted large-scale state transitions (pLST) status from CTC-derived
morphology features.Where the paper says it
--- Page 14 ---
QCML POVM achieves the highest ROC AUC score of (0.763) as shown in Table 2 and Figure 4 with ROC AUC curves for top performing models. QCML POVM is able to generate full probability densities of LST values as shown in Figure 5. p. 7 763) as shown in Table 2 and Figure 4 with ROC AUC curves for top
performing models. Additionally, QCML POVM is able to generate full probability densities of LST values as shown in Figure 5. 763) as shown in Table 2 and Figure 4 with ROC AUC curves for top
performing models. Additionally, QCML POVM is able to generate full probability densities of LST values as shown in Figure 5.Where the paper says it
Where the paper says it
--- Page 15 ---
QCML distance (a) QCML distance (b) Euclidean distance Using QCML distance (a) one can achieve a better separation between LST+ and stable CTCs.
--- Page 16 ---
QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI.
--- Page 17 ---
QCML gradient-based feature importance QCML identified cellular size, which is biologically relevant, as the presence of cytoplasmic micronuclei and multinucleation is permitted by larger-sized cells and is expected to occur in genomically unstable cells.
--- Page 18 ---
QCML gradient-based feature importance Fractal features are approximations of irregularity and complexity in cellular and nuclear shape, suggesting abnormal nuclear contours that are expected to occur during chromosomal instability.
--- Page 19 ---
QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI.
--- Page 20 ---
QCML gradient-based feature importance QCML identified cellular size, which is biologically relevant, as the presence of cytoplasmic micronuclei and multinucleation is permitted by larger-sized cells and is expected to occur in genomically unstable cells.
--- Page 21 ---
QCML gradient-based feature importance Fractal features are approximations of irregularity and complexity in cellular and nuclear shape, suggesting abnormal nuclear contours that are expected to occur during chromosomal instability.
--- Page 22 ---
QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI.
--- Page 23 ---
QCML gradient-based feature importance QCML identified cellular size, which is biologically relevant, as the presence of cytoplasmic micronuclei and multinucleation is permitted by larger-sized cells and is expected to occur in genomically unstable cells.
--- Page 24 ---
QCML gradient-based feature importance Fractal features are approximations of irregularity and complexity in cellular and nuclear shape, suggesting abnormal nuclear contours that are expected to occur during chromosomal instability.
--- Page 25 ---
QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI.
--- Page 26 ---
QCML gradient-based feature importance QCML identified cellular size, which is biologically relevant, as the presence of cytoplasmic micronuclei and multinucleation is permitted by larger-sized cells and is expected to occur in genomically unstable cells.
--- Page 27 ---
QCML gradient-based feature importance Fractal features are approximations of irregularity and complexity in cellular and nuclear shape, suggesting abnormal nuclear contours that are expected to occur during chromosomal instability.
--- Page 28 ---
QCML gradient-based feature importance The top 10 CTCs morphology and protein content features that the QCML classifier used as predictor of LST are measures of: (i) Protein correlation and colocalization between channels such as Cross-channel (LBP) for CK vs DAPI and cellular colocalization between CK-DAPI.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Quantum Cognition Machine Learning (QCML) can estimate morphology-predicted chromosomal instability in circulating tumor cells (CTCs) from patients with metastatic breast cancer by representing data as state vectors in a Hilbert space. This enables context-aware feature modeling, dimensionality reduction, and enhanced generalization without requiring curated feature selection.
-
The QCML Positive Operator-Valued Measure (POVM) allows the system to forecast the probability density function of a target variable instead of just point estimates. This capability enables the estimation of confidence intervals and provides probabilistic forecasts for whether chromosomal instability will be above or below a specified threshold, as shown in Figure 5.
-
The QCML model can utilize a mixed loss function incorporating L1 and cross-entropy components with varying weights on the positive class to achieve a desired balance between specificity and sensitivity. This allows the system to produce both real-valued LST forecasts and probability forecasts for LST being above or below a certain threshold.
-
The system can be trained using a gradient-based feature importance approach to rank CTC morphology features and protein expressions based on their impact on the final output, identifying features such as Cross-channel Local Binary Pattern (LBP) for CK vs DAPI as predictors of low chromosomal instability.
-
The QCML distance metric can be used to define a similarity measure between two data points based on quantum fidelity, which is much better at finding a separation between genomically unstable and stable CTCs compared to the standard Euclidean distance.
Abstract
The accurate prediction of chromosomal instability from the morphology of circulating tumor cells (CTCs) enables real-time detection of CTCs with high metastatic potential in the context of liquid biopsy diagnostics. However, it presents a significant challenge due to the high dimensionality and complexity of single-cell digital pathology data. Here, we introduce the application of Quantum Cognition Machine Learning (QCML), a quantum-inspired computational framework, to estimate morphology-predicted chromosomal instability in CTCs from patients with metastatic breast cancer. QCML leverages quantum mechanical principles to represent data as state vectors in a Hilbert space, enabling context-aware feature modeling, dimensionality reduction, and enhanced generalization without requiring curated feature selection. QCML outperforms conventional machine learning methods when tested on out of sample verification CTCs, achieving higher accuracy in identifying predicted large-scale state transitions (pLST) status from CTC-derived morphology features. These preliminary findings support the application of QCML as a novel machine learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts. QCML enables the simulation of cognition-like learning for the identification of biologically meaningful prediction of chromosomal instability from CTC morphology, offering a novel tool for CTC classification in liquid biopsy.
Sources
- Robust estimation of the intrinsic dimension of data sets with quantum cognition machine learning
- Supervised Similarity for High-Yield Corporate Bonds with Quantum Cognition Machine Learning
Related papers
- ViroGym: Realistic Large-Scale Benchmarks for Evaluating Viral Proteins
- CellCanon: An Agentic Cellular World Model for Prediction, Explanation, and Observation-Driven Updating
- Probing and steering biology across Boltz-1s trunk-diffusion boundary
- Trajectory Inference of Human Aging from Cross-Sectional DNA Methylation Data
- Bridging the Modality Bottleneck in Pathology MIL through Molecularly Guided Feature Adaptation
- Molecular notation can act as a shortcut in AI benchmark evaluation