Background: Quantum Machine Learning#

This section provides foundational information about quantum machine learning concepts and techniques used in QBioCode. This material can be used for seminars, presentations, and educational purposes.

Quantum Computing in Biology and Healthcare#

Quantum computing is emerging as a transformative technology for biological and healthcare applications:

Key Applications

Protein Structure Prediction: A perspective on protein structure prediction using quantum computers - J. Chem. Theory Comput. (2025).
Biomarker Discovery: How quantum computing can enhance biomarker discovery - Patterns (2025).
Single-Cell Omics: Advancing single-cell omics and cell-based therapeutics with quantum computing - Nature Reviews Molecular Cell Biology (2026).

These applications demonstrate the potential of quantum computing to address computationally challenging problems in structural biology, precision medicine, and cellular therapeutics, domains where QBioCode’s quantum machine learning tools can provide valuable insights.

Quantum Algorithms for Spatiotemporal Single-Cell Analysis#

The landscape of quantum machine learning algorithms for biological applications spans six broad computational tasks, each with classical methods that have quantum analogs. The figure below illustrates how these algorithms intersect to enable comprehensive spatiotemporal single-cell analysis:

Quantum machine learning algorithms for single-cell analysis — **Quantum algorithms landscape for spatiotemporal single-cell analysis.** Six broad computational tasks (colored lines) intersect with various applicable algorithms (circles), leading towards spatiotemporal single-cell analysis. Classical methods with quantum analogs are denoted with a preceding (Q). Algorithms include: QCNN (quantum convolutional neural network), QCumulant (quantum cumulant calculation), QGAN (quantum generative adversarial network), QGNN (quantum graph neural networks), QMCMC (quantum Monte Carlo Markov chain), QNN (quantum neural network), QODE (quantum ordinary differential equations), QTDA (quantum topological data analysis), QVAE (quantum variational autoencoders), and VQC (variational quantum circuit). Some methods require fault-tolerant quantum devices (FTQDs). *Source: Nature Reviews Molecular Cell Biology (2026).*#

Note

QBioCode Implementation Status: QBioCode currently implements several of these quantum algorithms including VQC, QNN, and quantum kernel methods (QSVC, PQK). As quantum hardware continues to advance, additional algorithms from this landscape will become practical for real-world biological applications.

Introduction to Quantum Computing#

Quantum computing leverages quantum mechanical phenomena such as superposition and entanglement to process information in fundamentally different ways than classical computers.

Key Concepts#

Qubits: The basic unit of quantum information
Superposition: A qubit can exist in multiple states simultaneously
Entanglement: Quantum states can be correlated in ways impossible classically
Quantum Gates: Operations that manipulate qubits

📺 Video Learning Resources

For comprehensive video tutorials on these quantum computing fundamentals, see the Qiskit YouTube Playlist: Understanding Quantum Information and Computation. This series provides excellent visual explanations of qubits, superposition, entanglement, and quantum gates.

Quantum Machine Learning#

Quantum Machine Learning (QML) combines quantum computing with machine learning algorithms to potentially achieve computational advantages for certain tasks.

🔍 Taxonomy of Machine Learning and Quantum Computing#

Understanding the relationship between machine learning and quantum computing requires considering both the type of data (classical or quantum) and the processing device (classical or quantum computer).

Reference

Maria Schuld, Francesco Petruccione: “Supervised Learning with Quantum Computers”, Page 6

💻 CC: Classical-Classical

Classical Data → Classical Processing

Machine Learning based on methods borrowed from Quantum Information research, or “Quantum-Inspired” algorithms.

Examples: Tensor networks, quantum-inspired optimization

🔬 CQ: Classical-Quantum

Classical Data → Quantum Processing

Synonym for QML. Data come from classical systems like text, images, time series, macro-economic variables.

Key Challenge: Requires Quantum-Classical interface

This is the primary focus of QBioCode

🤖 QC: Quantum-Classical

Quantum Data → Classical Processing

How can Machine Learning help with Quantum Computing?

Examples: Quantum error correction, quantum state tomography

⚛️ QQ: Quantum-Quantum

Quantum Data → Quantum Processing

Closely related to CQ. Data can be measured from quantum systems or datasets can be Quantum States.

Examples: Quantum state classification, quantum sensing

Important

CQ (Classical-Quantum) is the most relevant category for healthcare and life sciences applications, where classical biological data is processed using quantum algorithms to potentially achieve computational advantages.

🔬 Quantum Machine Learning in a Nutshell#

Both classical and quantum machine learning start with the same input (Dataset D) and produce the same output (Prediction ŷ), but the quantum approach introduces two critical intermediate steps:

Classical Machine Learning

Workflow:

[Dataset D] → ML Algorithm → [Prediction ŷ]

Traditional approach: data flows directly through classical algorithms.

Quantum Machine Learning

Workflow:

[Dataset D] → 🔵 Encoding → Quantum Circuit → 🔵 Measurement → [Prediction ŷ]

Two Critical Quantum Steps:

🔵 Encoding: Transforms classical data into quantum states (e.g., qubit rotations, amplitude encoding)

Maps classical features to quantum parameters
Enables quantum superposition and entanglement
Foundation for quantum advantage

🔵 Measurement: Extracts classical information from quantum states

Collapses quantum superposition to classical bits
Produces probabilistic outcomes
Bridges quantum and classical worlds

Important

Key Insight: The quantum workflow shares the same endpoints (classical data in, predictions out) but leverages quantum phenomena in between. The Encoding and Measurement steps are what make quantum machine learning fundamentally different, and potentially more powerful, than classical approaches.

Key References

M. Schuld, F. Petruccione: “Supervised Learning with Quantum Computers”
M. Schuld, N. Killoran: arXiv 1803.07128

🔑 Data Encoding: The Foundation of QML#

Important

Data encoding is one of the most important parts of QML. Frameworks, software, and hardware that address the interface between classical memory and quantum devices are key for runtime evaluations.

Encoding Strategies#

📊 Qubit-Efficient State Preparation

Encode data in superposition states
Minimize the number of qubits required
Optimal for limited qubit resources

📈 Amplitude-Efficient State Preparation

Encode information in quantum state amplitudes
Leverage quantum parallelism
Efficient for high-dimensional data

Encoding Methods: Complexity Analysis#

Different encoding methods have different resource requirements in terms of qubits and state preparation time.

Notation:

N = number of features
n = number of qubits

Encoding Method Comparison

Encoding Method	# Qubits	State Prep Runtime	Description
Basis	nN	O(N)	Direct mapping of classical bits to quantum basis states
Amplitude	log(N)	O(N) to O(log N)	Encode data in quantum state amplitudes (most qubit-efficient)
Angle	N	O(N)	Encode data in rotation angles of quantum gates
Arbitrary	n	O(N)	General feature map encoding to high-dimensional spaces

Note

Key Insight: Amplitude encoding is the most qubit-efficient method, requiring only log(N) qubits for N features, but may require more complex state preparation circuits. The choice of encoding method depends on the specific application and available quantum hardware.

🗺️ Quantum Feature Maps#

Quantum feature maps are fundamental to quantum machine learning, as they define how classical data is encoded into quantum states. The structure of the feature map directly impacts the expressiveness and performance of quantum algorithms, enabling quantum systems to exploit higher-dimensional Hilbert spaces efficiently and create decision boundaries that are difficult for classical kernel functions to replicate.

Why Quantum Feature Maps Matter#

The power of quantum feature maps lies in their ability to transform classical data through quantum operations, creating complex decision boundaries adapted to the data’s inherent structure. By controlling rotation factors (α) and Pauli gate combinations, quantum kernels can:

Adapt to data complexity: Simple boundaries for straightforward patterns, complex boundaries for intricate data
Capture feature interactions: Through entangling operations that create correlations in the quantum state
Avoid overfitting: By tuning the expressiveness to match the problem’s inherent complexity

QSVM Decision Boundaries — **Adaptive Decision Boundaries with Quantum Feature Maps:** Demonstration of quantum SVM flexibility across different data patterns. **(a)** XOR-patterned data requiring simple separating boundaries. **(b)** Complex data with intricate class distributions. **(c)** QSVM decision boundary effectively separating XOR patterns. **(d)** QSVM decision boundary adapting to complex data structure. The same quantum kernel framework can handle both simple and complex classification tasks by adjusting feature map parameters. Source: Park et al. (2020), arXiv:2012.07725#

Mathematical Foundation#

The quantum feature map is constructed using parameterized unitary transformations:

\[U_\varphi(\mathbf{x}) = \exp\left(i \sum_{j=1}^{n} \alpha_j \varphi_s(x_j) \prod_{\sigma_j \in \{X,Y,Z\}} \sigma_j\right)\]

where:

$\alpha_j$ : Rotation factors controlling phase rotation based on feature values
$\sigma_j$ : Pauli operators ($X, Y, Z$) defining the rotation axes
$\varphi_s(\mathbf{x})$ : Scaled classical features (typically mapped to $[0, 2\pi]$)
$n$ : Number of features

This formulation enables:

Ising-like interactions (ZZ, YY gates) for feature coupling:

\[U_{ZZ}(\theta) = \exp(-i\theta Z \otimes Z)\]

Independent rotations (single Z, Y gates) for separable features:

\[R_Z(\theta) = \exp(-i\theta Z/2), \quad R_Y(\theta) = \exp(-i\theta Y/2)\]

Flexible hyperplanes adapted to data patterns through hyperparameter optimization of $\alpha_j$ and gate combinations.

Feature Map Architectures#

Different feature map designs offer varying levels of complexity and entanglement. Understanding these architectures helps in selecting the appropriate approach for specific problems:

🔄 Independent Rotations

Single-qubit rotations applied independently to each feature, creating straightforward decision surfaces.

Characteristics:

Simple circuit structure
No entanglement
Limited expressiveness
Fast execution

Best for: XOR patterns, linearly separable data, baseline comparisons

🔗 Entangled Rotations

Two-qubit gates creating quantum correlations between features, capturing feature dependencies.

Characteristics:

Introduces entanglement
Captures feature interactions
Higher expressiveness
More complex circuits

Best for: Data with feature dependencies and correlations

🎯 Hybrid Approach

Combination of independent and entangling operations, creating higher-order decision boundaries.

Characteristics:

Balanced complexity
Flexible design
Tunable expressiveness
Practical for current hardware

Best for: Real-world biological data, most QML applications

Choosing a Feature Map#

The choice of feature map depends on several factors:

Problem Structure: Does the problem benefit from capturing feature interactions?
Hardware Constraints: Available qubits, gate fidelity, and circuit depth limits
Data Characteristics: Dimensionality, correlations, and noise levels
Computational Resources: Classical preprocessing and quantum execution time

Important

Key Insight: The same quantum kernel framework can be tuned from simple to complex by adjusting rotation factors and Pauli gate combinations. This flexibility allows quantum algorithms to match the complexity of the data without creating unnecessarily complex decision boundaries.

Note

In QBioCode, feature maps can be configured through the encoding parameters in quantum model functions (e.g., compute_qsvc, compute_pqk), allowing users to experiment with different architectures for their specific applications.

Reference

Park, J.-E., Quanz, B., Wood, S., Higgins, H., & Harishankar, R. (2020). Practical application improvement to Quantum SVM: theory to practice. arXiv:2012.07725. https://arxiv.org/abs/2012.07725

💡 Why Quantum Machine Learning?#

⚡ Computational Advantage

Potential speedups for specific tasks through quantum parallelism and interference

📊 Sample Efficiency

Reduced training data requirements through quantum generalization properties

🌐 High-Dimensional Spaces

Natural representation of complex data through quantum encoding in Hilbert spaces

🔬 Novel Algorithms

New approaches to classification and optimization leveraging quantum properties

Understanding Quantum Advantage#

Quantum advantage in machine learning can manifest in multiple ways, each with distinct theoretical foundations and practical implications. Importantly, quantum advantage is not solely about computational speed, it encompasses accuracy improvements, sample efficiency, and empirical verifiability.

Beyond Speed: A Multifaceted Advantage

Recent work has demonstrated that quantum advantage should be evaluated across multiple dimensions:

Computational Speed: Reducing time complexity through quantum parallelism and interference
Prediction Accuracy: Achieving higher classification or regression accuracy than classical methods
Sample Efficiency: Learning from fewer training examples while maintaining performance
Empirical Verifiability: Demonstrating measurable improvements on real-world problems

Key Reference: Redefining Quantum Advantage

Huang, H.-Y. et al. (2024). “Quantum advantage in learning from experiments.” arXiv:2506.20658. https://arxiv.org/abs/2506.20658

This work emphasizes that quantum advantage should be empirically verifiable and not limited to computational speedup. Quantum algorithms can provide advantages through improved accuracy and learning efficiency, even when classical algorithms exist with similar time complexity.

Computational Complexity Advantage

Traditional quantum advantage focuses on computational speedup, reducing the time required to solve a problem. Quantum algorithms can potentially offer exponential or polynomial speedups for specific tasks through quantum parallelism, superposition, and interference effects.

Additional References

Optimization Challenges: Abbas, A. et al. (2024). “Challenges and opportunities in quantum optimization.” Nature Reviews Physics, 6, 718–735.
QML Landscape: Cerezo, M. et al. (2022). “Challenges and opportunities in quantum machine learning.” Nature Computational Science, 2, 567–576.

Accuracy Advantage

Quantum machine learning models can achieve higher prediction accuracy than classical counterparts on certain problems, even when computational time is comparable. This accuracy advantage stems from:

Richer Feature Spaces: Quantum feature maps can capture complex nonlinear relationships that are difficult for classical kernels
Quantum Interference: Constructive and destructive interference can enhance signal-to-noise ratios
Entanglement: Correlations between quantum states can encode higher-order feature interactions

Important

Empirical Verification is Essential

Quantum advantage claims must be empirically verifiable on real-world problems, not just theoretical constructs. This requires:

Benchmarking against state-of-the-art classical methods
Testing on diverse datasets with varying characteristics
Accounting for hardware noise and error rates
Demonstrating reproducibility across multiple runs

Sample Complexity Advantage

Beyond computational speed, quantum learning theory introduces another dimension of advantage: sample complexity, the number of training examples needed to achieve a target accuracy. This is particularly relevant within the Probably Approximately Correct (PAC) learning framework.

In this context, quantum advantage emerges when a quantum learning algorithm requires fewer training samples than any classical algorithm to reach the same prediction accuracy. Specifically, if the upper bound on sample complexity for a quantum learner is lower than the lower bound for all classical learners, a provable quantum advantage exists.

Important

Sample Efficiency vs. Computational Efficiency

While low sample complexity is crucial for data-limited scenarios, it doesn’t automatically guarantee practical efficiency. The computational cost of identifying useful training examples or processing them may still be substantial. True practical advantage requires considering both sample and computational complexity together.

Generalization from Few Training Data

Recent theoretical and empirical work has demonstrated that quantum machine learning models can exhibit superior generalization properties, learning effectively from remarkably small training sets. This capability is particularly valuable in domains where:

Data acquisition is expensive (e.g., quantum experiments, medical imaging)
Labeling requires expert knowledge (e.g., drug discovery, materials science)
Privacy constraints limit data availability (e.g., healthcare, finance)

Key Reference

Caro, M. C. et al. (2022). “Generalization in quantum machine learning from few training data.” Nature Communications, 13, 4919.

Theoretical Framework: PAC Learning

The PAC learning framework provides a rigorous foundation for analyzing sample complexity:

Goal: Learn a hypothesis that is “probably” (with high probability) “approximately correct” (within error threshold $\epsilon$)
Sample Complexity: Number of training examples $m$ needed to achieve error $\leq \epsilon$ with probability $\geq 1-\delta$
Quantum Advantage: When quantum algorithms achieve this with $m_{\text{quantum}} \ll m_{\text{classical}}$

Key Reference

Arunachalam, S. & de Wolf, R. (2017). “Guest column: A survey of quantum learning theory.” ACM SIGACT News, 48(2), 41-67.

Practical Considerations

While theoretical advantages are promising, practical quantum advantage depends on:

Hardware Quality: Gate fidelity, coherence times, and qubit connectivity
Problem Structure: Not all problems benefit equally from quantum approaches
Data Characteristics: Quantum advantage may be problem-specific
Hybrid Approaches: Combining quantum and classical methods often yields best results

Note

Current Status: With current quantum hardware, demonstrating practical quantum advantage for real-world machine learning tasks remains an active area of research. QBioCode provides tools to explore these possibilities empirically across both quantum simulators and available quantum devices.

QML Algorithms in QBioCode

QBioCode provides implementations of state-of-the-art quantum machine learning algorithms alongside classical baselines for comprehensive benchmarking. These quantum methods address key challenges in classical approaches, such as computational expense, poor out-of-distribution generalization, and sensitivity to noisy data, by leveraging quantum properties like superposition and entanglement.

Quantum Advantage in Biological Applications

Quantum machine learning methods show particular promise for biological data analysis, including cell classification and segmentation, trajectory inference, perturbation modeling, and multi-omics integration. These applications benefit from quantum algorithms’ ability to handle high-dimensional data and learn from limited samples.

For detailed API documentation, see the API Overview.

Quantum Kernels

Quantum kernels compute similarity measures between data points in high-dimensional Hilbert spaces, offering advantages over classical kernels for complex pattern recognition.

Implementations:

Projected Quantum Kernel (PQK) - compute_pqk: Projects data into quantum feature space for enhanced separability in classification tasks.
Quantum Support Vector Classifier (QSVC) - compute_qsvc: Quantum analog of SVM using quantum kernels for binary and multi-class classification.

Note

Biological Applications: Quantum kernels have shown promise in cell classification, CpG prediction, and T cell receptor classification, where they can capture complex nonlinear relationships in biological data.

Variational Quantum Algorithms

Variational quantum algorithms use parameterized quantum circuits optimized through hybrid quantum-classical loops, analogous to neural networks but leveraging quantum computational advantages.

Implementations:

Variational Quantum Classifier (VQC) - compute_vqc: Trainable quantum circuit for classification with gradient-based optimization.
Quantum Neural Networks (QNN) - compute_qnn: Quantum analog of neural networks with parameterized quantum gates as trainable layers.

Note

Biological Applications: Quantum variational autoencoders have been applied to trajectory inference, perturbation modeling, and multi-omics integration, addressing challenges like learning uninformative latent representations and sensitivity to noisy data that affect classical variational autoencoders.

Classical Baselines

QBioCode provides classical machine learning models for rigorous benchmarking and comparison:

Support Vector Classifier (SVC) - compute_svc: Classical kernel-based classifier for comparison with quantum kernels.
Random Forest (RF) - compute_rf: Ensemble method combining multiple decision trees for robust predictions.
Logistic Regression (LR) - compute_lr: Linear model providing interpretable baseline for binary classification.
Decision Tree (DT) - compute_dt: Tree-based model offering interpretable decision rules.
Naive Bayes (NB) - compute_nb: Probabilistic classifier based on Bayes’ theorem with feature independence assumption.
Multi-layer Perceptron (MLP) - compute_mlp: Classical neural network for comparison with quantum neural networks.

Important

Benchmarking Strategy: Comparing quantum algorithms against these classical baselines is essential for demonstrating quantum advantage. QBioCode’s unified interface enables fair comparisons across all methods using identical data preprocessing and evaluation metrics.