A spectroscopic liquid biopsy for the earlier detection of multiple cancer types.

BACKGROUND: A rapid, low-cost blood test that can be applied to reliably detect multiple different cancer types would be transformational. METHODS: In this large-scale discovery study (n = 2092 patients) we applied the Dxcover® Cancer Liquid Biopsy to examine eight different cancers. The test uses Fourier transform infrared (FTIR) spectroscopy and machine-learning algorithms to detect cancer. RESULTS: Area under the receiver operating characteristic curve (ROC) values were calculated for eight cancer types versus symptomatic non-cancer controls: brain (0.90), breast (0.76), colorectal (0.91), kidney (0.91), lung (0.91), ovarian (0.86), pancreatic (0.84) and prostate (0.86). We assessed the test performance when all eight cancer types were pooled to classify 'any cancer' against non-cancer patients. The cancer versus asymptomatic non-cancer classification detected 64% of Stage I cancers when specificity was 99% (overall sensitivity 57%). When tuned for higher sensitivity, this model identified 99% of Stage I cancers (with specificity 59%). CONCLUSIONS: This spectroscopic blood test can effectively detect early-stage disease and can be fine-tuned to maximise either sensitivity or specificity depending on the requirements from different healthcare systems and cancer diagnostic pathways. This low-cost strategy could facilitate the requisite earlier diagnosis, when cancer treatment can be more effective, or less toxic. STATEMENT OF TRANSLATIONAL RELEVANCE: The earlier diagnosis of cancer is of paramount importance to improve patient survival. Current liquid biopsies are mainly focused on single tumour-derived biomarkers, which limits test sensitivity, especially for early-stage cancers that do not shed enough genetic material. This pan-omic liquid biopsy analyses the full complement of tumour and immune-derived markers present within blood derivatives and could facilitate the earlier detection of multiple cancer types. There is a low barrier to integrating this blood test into existing diagnostic pathways since the technology is rapid, simple to use, only minute sample volumes are required, and sample preparation is minimal. In addition, the spectroscopic liquid biopsy described in this study has the potential to be combined with other orthogonal tests, such as cell-free DNA, which could provide an efficient route to diagnosis. Cancer treatment can be more effective when given earlier, and this low-cost strategy has the potential to improve patient prognosis.

Liquid biopsies: the future of cancer early detection.

Cancer is a worldwide pandemic. The burden it imposes grows steadily on a global scale causing emotional, physical, and financial strains on individuals, families, and health care systems. Despite being the second leading cause of death worldwide, many cancers do not have screening programs and many people with a high risk of developing cancer fail to follow the advised medical screening regime due to the nature of the available screening tests and other challenges with compliance. Moreover, many liquid biopsy strategies being developed for early detection of cancer lack the sensitivity required to detect early-stage cancers. Early detection is key for improved quality of life, survival, and to reduce the financial burden of cancer treatments which are greater at later stage detection. This review examines the current liquid biopsy market, focusing in particular on the strengths and drawbacks of techniques in achieving early cancer detection. We explore the clinical utility of liquid biopsy technologies for the earlier detection of solid cancers, with a focus on how a combination of various spectroscopic and -omic methodologies may pave the way for more efficient cancer diagnostics.

Augmentation of FTIR spectral datasets using Wasserstein generative adversarial networks for cancer liquid biopsies.

Over recent years, deep learning (DL) has become more widely used within the field of cancer diagnostics. However, DL often requires large training datasets to prevent overfitting, which can be difficult and expensive to acquire. Data augmentation is a method that can be used to generate new data points to train DL models. In this study, we use attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectra of patient dried serum samples and compare non-generative data augmentation methods to Wasserstein generative adversarial networks (WGANs) in their ability to improve the performance of a convolutional neural network (CNN) to differentiate between pancreatic cancer and non-cancer samples in a total cohort of 625 patients. The results show that WGAN augmented spectra improve CNN performance more than non-generative augmented spectra. When compared with a model that utilised no augmented spectra, adding WGAN augmented spectra to a CNN with the same architecture and same parameters, increased the area under the receiver operating characteristic curve (AUC) from 0.661 to 0.757, presenting a 15% increase in diagnostic performance. In a separate test on a colorectal cancer dataset, data augmentation using a WGAN led to an increase in AUC from 0.905 to 0.955. This demonstrates the impact data augmentation can have on DL performance for cancer diagnosis when the amount of real data available for model training is limited.

Recurrent neural networks for time domain modelling of FTIR spectra: application to brain tumour detection.

Attenuated total reflectance (ATR)-Fourier transform infrared (FTIR) spectroscopy alongside machine learning (ML) techniques is an emerging approach for the early detection of brain cancer in clinical practice. A crucial step in the acquisition of an IR spectrum is the transformation of the time domain signal from the biological sample to a frequency domain spectrum via a discrete Fourier transform. Further pre-processing of the spectrum is typically applied to reduce non-biological sample variance, and thus to improve subsequent analysis. However, the Fourier transformation is often assumed to be essential even though modelling of time domain data is common in other fields. We apply an inverse Fourier transform to frequency domain data to map these to the time domain. We use the transformed data to develop deep learning models utilising Recurrent Neural Networks (RNNs) to differentiate between brain cancer and control in a cohort of 1438 patients. The best performing model achieves a mean (cross-validated score) area under the receiver operating characteristic (ROC) curve (AUC) of 0.97 with sensitivity of 0.91 and specificity of 0.91. This is better than the optimal model trained on frequency domain data which achieves an AUC of 0.93 with sensitivity of 0.85 and specificity of 0.85. A dataset comprising 385 patient samples which were prospectively collected in the clinic is used to test a model defined with the best performing configuration and fit to the time domain. Its classification accuracy is found to be comparable to the gold-standard for this dataset demonstrating that RNNs can accurately classify disease states using spectroscopic data represented in the time domain.

Blinded Predictions and Post Hoc Analysis of the Second Solubility Challenge Data: Exploring Training Data and Feature Set Selection for Machine and Deep Learning Models.

Accurate methods to predict solubility from molecular structure are highly sought after in the chemical sciences. To assess the state of the art, the American Chemical Society organized a "Second Solubility Challenge" in 2019, in which competitors were invited to submit blinded predictions of the solubilities of 132 drug-like molecules. In the first part of this article, we describe the development of two models that were submitted to the Blind Challenge in 2019 but which have not previously been reported. These models were based on computationally inexpensive molecular descriptors and traditional machine learning algorithms and were trained on a relatively small data set of 300 molecules. In the second part of the article, to test the hypothesis that predictions would improve with more advanced algorithms and higher volumes of training data, we compare these original predictions with those made after the deadline using deep learning models trained on larger solubility data sets consisting of 2999 and 5697 molecules. The results show that there are several algorithms that are able to obtain near state-of-the-art performance on the solubility challenge data sets, with the best model, a graph convolutional neural network, resulting in an RMSE of 0.86 log units. Critical analysis of the models reveals systematic differences between the performance of models using certain feature sets and training data sets. The results suggest that careful selection of high quality training data from relevant regions of chemical space is critical for prediction accuracy but that other methodological issues remain problematic for machine learning solubility models, such as the difficulty in modeling complex chemical spaces from sparse training data sets.

Solvation entropy, enthalpy and free energy prediction using a multi-task deep learning functional in 1D-RISM.

Simultaneous calculation of entropies, enthalpies and free energies has been a long-standing challenge in computational chemistry, partly because of the difficulty in obtaining estimates of all three properties from a single consistent simulation methodology. This has been particularly true for methods from the Integral Equation Theory of Molecular Liquids such as the Reference Interaction Site Model which have traditionally given large errors in solvation thermodynamics. Recently, we presented pyRISM-CNN, a combination of the 1 Dimensional Reference Interaction Site Model (1D-RISM) solver, pyRISM, with a deep learning based free energy functional, as a method of predicting solvation free energy (SFE). With this approach, a 40-fold improvement in prediction accuracy was delivered for a multi-solvent, multi-temperature dataset when compared to the standard 1D-RISM theory [Fowles et al., Digital Discovery, 2023, 2, 177-188]. Here, we report three further developments to the pyRISM-CNN methodology. Firstly, solvation free energies have been introduced for organic molecular ions in methanol or water solvent systems at 298 K, with errors below 4 kcal mol-1 obtained without the need for corrections or additional descriptors. Secondly, the number of solvents in the training data has been expanded from carbon tetrachloride, water and chloroform to now also include methanol. For neutral solutes, prediction errors nearing or below 1 kcal mol-1 are obtained for each organic solvent system at 298 K and water solvent systems at 273-373 K. Lastly, pyRISM-CNN was successfully applied to the simultaneous prediction of solvation enthalpy, entropy and free energy through a multi-task learning approach, with errors of 1.04, 0.98 and 0.47 kcal mol-1, respectively, for water solvent systems at 298 K.

Early economic evaluation to guide the development of a spectroscopic liquid biopsy for the detection of brain cancer.

OBJECTIVES: An early economic evaluation to inform the translation into clinical practice of a spectroscopic liquid biopsy for the detection of brain cancer. Two specific aims are (1) to update an existing economic model with results from a prospective study of diagnostic accuracy and (2) to explore the potential of brain tumor-type predictions to affect patient outcomes and healthcare costs. METHODS: A cost-effectiveness analysis from a UK NHS perspective of the use of spectroscopic liquid biopsy in primary and secondary care settings, as well as a cost-consequence analysis of the addition of tumor-type predictions was conducted. Decision tree models were constructed to represent simplified diagnostic pathways. Test diagnostic accuracy parameters were based on a prospective validation study. Four price points (GBP 50-200, EUR 57-228) for the test were considered. RESULTS: In both settings, the use of liquid biopsy produced QALY gains. In primary care, at test costs below GBP 100 (EUR 114), testing was cost saving. At GBP 100 (EUR 114) per test, the ICER was GBP 13,279 (EUR 15,145), whereas at GBP 200 (EUR 228), the ICER was GBP 78,300 (EUR 89,301). In secondary care, the ICER ranged from GBP 11,360 (EUR 12,956) to GBP 43,870 (EUR 50,034) across the range of test costs. CONCLUSIONS: The results demonstrate the potential for the technology to be cost-effective in both primary and secondary care settings. Additional studies of test use in routine primary care practice are needed to resolve the remaining issues of uncertainty-prevalence in this patient population and referral behavior.

Exploring Ligand Stability in Protein Crystal Structures Using Binding Pose Metadynamics.

Identification of correct protein-ligand binding poses is important in structure-based drug design and crucial for the evaluation of protein-ligand binding affinity. Protein-ligand coordinates are commonly obtained from crystallography experiments that provide a static model of an ensemble of conformations. Binding pose metadynamics (BPMD) is an enhanced sampling method that allows for an efficient assessment of ligand stability in solution. Ligand poses that are unstable under the bias of the metadynamics simulation are expected to be infrequently occupied in the energy landscape, thus making minimal contributions to the binding affinity. Here, the robustness of the method is studied using crystal structures with ligands known to be incorrectly modeled, as well as 63 structurally diverse crystal structures with ligand fit to electron density from the Twilight database. Results show that BPMD can successfully differentiate compounds whose binding pose is not supported by the electron density from those with well-defined electron density.

Developing infrared spectroscopic detection for stratifying brain tumour patients: glioblastoma multiforme vs. lymphoma.

Over a third of brain tumour patients visit their general practitioner more than five times prior to diagnosis in the UK, leading to 62% of patients being diagnosed as emergency presentations. Unfortunately, symptoms are non-specific to brain tumours, and the majority of these patients complain of headaches on multiple occasions before being referred to a neurologist. As there are currently no methods in place for the early detection of brain cancer, the affected patients' average life expectancy is reduced by 20 years. These statistics indicate that the current pathway is ineffective, and there is a vast need for a rapid diagnostic test. Attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectroscopy is sensitive to the hallmarks of cancer, as it analyses the full range of macromolecular classes. The combination of serum spectroscopy and advanced data analysis has previously been shown to rapidly and objectively distinguish brain tumour severity. Recently, a novel high-throughput ATR accessory has been developed, which could be cost-effective to the National Health Service in the UK, and valuable for clinical translation. In this study, 765 blood serum samples have been collected from healthy controls and patients diagnosed with various types of brain cancer, contributing to one of the largest spectroscopic studies to date. Three robust machine learning techniques - random forest, partial least squares-discriminant analysis and support vector machine - have all provided promising results. The novel high-throughput technology has been validated by separating brain cancer and non-cancer with balanced accuracies of 90% which is comparable to the traditional fixed diamond crystal methodology. Furthermore, the differentiation of brain tumour type could be useful for neurologists, as some are difficult to distinguish through medical imaging alone. For example, the highly aggressive glioblastoma multiforme and primary cerebral lymphoma can appear similar on magnetic resonance imaging (MRI) scans, thus are often misdiagnosed. Here, we report the ability of infrared spectroscopy to distinguish between glioblastoma and lymphoma patients, at a sensitivity and specificity of 90.1% and 86.3%, respectively. A reliable serum diagnostic test could avoid the need for surgery and speed up time to definitive chemotherapy and radiotherapy.

On the effect of mutations in bovine or camel chymosin on the thermodynamics of binding κ-caseins.

Bovine and camel chymosins are aspartic proteases that are used in dairy food manufacturing. Both enzymes catalyze proteolysis of a milk protein, κ-casein, which helps to initiate milk coagulation. Surprisingly, camel chymosin shows a 70% higher clotting activity than bovine chymosin for bovine milk, while exhibiting only 20% of the unspecific proteolytic activity. By contrast, bovine chymosin is a poor coagulant for camel milk. Although both enzymes are marketed commercially, the disparity in their catalytic activity is not yet well understood at a molecular level, due in part to a lack of atomistic resolution data about the chymosin-κ-casein complexes. Here, we report computational alanine scanning calculations of all four chymosin-κ-casein complexes, allowing us to elucidate the influence that individual residues have on binding thermodynamics. Of the 12 sequence differences in the binding sites of bovine and camel chymosin, eight are shown to be particularly important for understanding differences in the binding thermodynamics (Asp112Glu, Lys221Val, Gln242Arg, Gln278Lys. Glu290Asp, His292Asn, Gln294Glu, and Lys295Leu. Residue in bovine chymosin written first). The relative binding free energies of single-point mutants of chymosin are calculated using the molecular mechanics three dimensional reference interaction site model (MM-3DRISM). Visualization of the solvent density functions calculated by 3DRISM reveals the difference in solvation of the binding sites of chymosin mutants.

Optimised spectral pre-processing for discrimination of biofluids via ATR-FTIR spectroscopy.

Pre-processing is an essential step in the analysis of spectral data. Mid-IR spectroscopy of biological samples is often subject to instrumental and sample specific variances which may often conceal valuable biological information. Whilst pre-processing can effectively reduce this unwanted variance, the plethora of possible processing steps has resulted in a lack of consensus in the field, often meaning that analysis outputs are not comparable. As pre-processing is specific to the sample under investigation, here we present a systematic approach for defining the optimum pre-processing protocol for biofluid ATR-FTIR spectroscopy. Using a trial-and-error based approach and a clinically relevant dataset describing control and brain cancer patients, the effects of pre-processing permutations on subsequent classification algorithms were observed, by assessing key diagnostic performance parameters, including sensitivity and specificity. It was found that optimum diagnostic performance correlated with the use of minimal binning and baseline correction, with derivative functions improving diagnostic performance most significantly. If smoothing is required, a Sovitzky-Golay approach was the preferred option in this investigation. Heavy binning appeared to reduce classification most significantly, alongside wavelet noise reduction (filter length ≥6), resulting in the lowest diagnostic performances of all pre-processing permutations tested.

Comparative Molecular Field Analysis Using Molecular Integral Equation Theory.

Recently, Güssregen et al. used solute-solvent distribution functions calculated by the three-dimensional Reference Interaction Site Model (3DRISM) in a 3D quantitative structure-activity relationship (QSAR) approach to model activity data for a set of serine protease inhibitors; this approach was referred to as Comparative Analysis of 3D RISM Maps (CARMa). [ J. Chem. Inf. MODEL: 2017, 57, 1652-1666] Here we extend this idea by introducing probe atoms into the 3DRISM solvent model in order to directly capture other molecular interactions in addition to those related to hydration/dehydration. Benchmark results for six different protein-ligand systems show that CARMa models trained on probe atom descriptors give consistently more accurate predictions than Comparative Molecular Field Analysis (CoMFA) and other common QSAR approaches.

Combining random forest and 2D correlation analysis to identify serum spectral signatures for neuro-oncology.

Fourier transform infrared (FTIR) spectroscopy has long been established as an analytical technique for the measurement of vibrational modes of molecular systems. More recently, FTIR has been used for the analysis of biofluids with the aim of becoming a tool to aid diagnosis. For the clinician, this represents a convenient, fast, non-subjective option for the study of biofluids and the diagnosis of disease states. The patient also benefits from this method, as the procedure for the collection of serum is much less invasive and stressful than traditional biopsy. This is especially true of patients in whom brain cancer is suspected. A brain biopsy is very unpleasant for the patient, potentially dangerous and can occasionally be inconclusive. We therefore present a method for the diagnosis of brain cancer from serum samples using FTIR and machine learning techniques. The scope of the study involved 433 patients from whom were collected 9 spectra each in the range 600-4000 cm(-1). To begin the development of the novel method, various pre-processing steps were investigated and ranked in terms of final accuracy of the diagnosis. Random forest machine learning was utilised as a classifier to separate patients into cancer or non-cancer categories based upon the intensities of wavenumbers present in their spectra. Generalised 2D correlational analysis was then employed to further augment the machine learning, and also to establish spectral features important for the distinction between cancer and non-cancer serum samples. Using these methods, sensitivities of up to 92.8% and specificities of up to 91.5% were possible. Furthermore, ratiometrics were also investigated in order to establish any correlations present in the dataset. We show a rapid, computationally light, accurate, statistically robust methodology for the identification of spectral features present in differing disease states. With current advances in IR technology, such as the development of rapid discrete frequency collection, this approach is of importance to enable future clinical translation and enables IR to achieve its potential.

Are the Sublimation Thermodynamics of Organic Molecules Predictable?

We compare a range of computational methods for the prediction of sublimation thermodynamics (enthalpy, entropy, and free energy of sublimation). These include a model from theoretical chemistry that utilizes crystal lattice energy minimization (with the DMACRYS program) and quantitative structure property relationship (QSPR) models generated by both machine learning (random forest and support vector machines) and regression (partial least squares) methods. Using these methods we investigate the predictability of the enthalpy, entropy and free energy of sublimation, with consideration of whether such a method may be able to improve solubility prediction schemes. Previous work has suggested that the major source of error in solubility prediction schemes involving a thermodynamic cycle via the solid state is in the modeling of the free energy change away from the solid state. Yet contrary to this conclusion other work has found that the inclusion of terms such as the enthalpy of sublimation in QSPR methods does not improve the predictions of solubility. We suggest the use of theoretical chemistry terms, detailed explicitly in the Methods section, as descriptors for the prediction of the enthalpy and free energy of sublimation. A data set of 158 molecules with experimental sublimation thermodynamics values and some CSD refcodes has been collected from the literature and is provided with their original source references.

Salting-out effects by pressure-corrected 3D-RISM.

We demonstrate that using a pressure corrected three-dimensional reference interaction site model one can accurately predict salting-out (Setschenow's) constants for a wide range of organic compounds in aqueous solutions of NaCl. The approach, based on classical molecular force fields, offers an alternative to more heavily parametrized methods.

Fast and General Method To Predict the Physicochemical Properties of Druglike Molecules Using the Integral Equation Theory of Molecular Liquids.

We report a method to predict physicochemical properties of druglike molecules using a classical statistical mechanics based solvent model combined with machine learning. The RISM-MOL-INF method introduced here provides an accurate technique to characterize solvation and desolvation processes based on solute-solvent correlation functions computed by the 1D reference interaction site model of the integral equation theory of molecular liquids. These functions can be obtained in a matter of minutes for most small organic and druglike molecules using existing software (RISM-MOL) (Sergiievskyi, V. P.; Hackbusch, W.; Fedorov, M. V. J. Comput. Chem. 2011, 32, 1982-1992). Predictions of caco-2 cell permeability and hydration free energy obtained using the RISM-MOL-INF method are shown to be more accurate than the state-of-the-art tools for benchmark data sets. Due to the importance of solvation and desolvation effects in biological systems, it is anticipated that the RISM-MOL-INF approach will find many applications in biophysical and biomedical property prediction.

Communication: Accurate hydration free energies at a wide range of temperatures from 3D-RISM.

We present a new model for computing hydration free energies by 3D reference interaction site model (3D-RISM) that uses an appropriate initial state of the system (as suggested by Sergiievskyi et al.). The new adjustment to 3D-RISM theory significantly improves hydration free energy predictions for various classes of organic molecules at both ambient and non-ambient temperatures. An extensive benchmarking against experimental data shows that the accuracy of the model is comparable to (much more computationally expensive) molecular dynamics simulations. The calculations can be readily performed with a standard 3D-RISM algorithm. In our work, we used an open source package AmberTools; a script to automate the whole procedure is available on the web (https://github.com/MTS-Strathclyde/ISc).

Structural interrogation of phosphoproteome identified by mass spectrometry reveals allowed and disallowed regions of phosphoconformation.

BACKGROUND: High-throughput mass spectrometric (HT-MS) study is the method of choice for monitoring global changes in proteome. Data derived from these studies are meant for further validation and experimentation to discover novel biological insights. Here we evaluate use of relative solvent accessible surface area (rSASA) and DEPTH as indices to assess experimentally determined phosphorylation events deposited in PhosphoSitePlus. RESULTS: Based on accessibility, we map these identifications on allowed (accessible) or disallowed (inaccessible) regions of phosphoconformation. Surprisingly a striking number of HT-MS/MS derived events (1461/5947 sites or 24.6%) are present in the disallowed region of conformation. By considering protein dynamics, autophosphorylation events and/or the sequence specificity of kinases, 13.8% of these phosphosites can be moved to the allowed region of conformation. We also demonstrate that rSASA values can be used to increase the confidence of identification of phosphorylation sites within an ambiguous MS dataset. CONCLUSION: While MS is a stand-alone technique for the identification of vast majority of phosphorylation events, identifications within disallowed region of conformation will benefit from techniques that independently probe for phosphorylation and protein dynamics. Our studies also imply that trapping alternate protein conformations may be a viable alternative to the design of inhibitors against mutation prone drug resistance kinases.

Is experimental data quality the limiting factor in predicting the aqueous solubility of druglike molecules?

We report the results of testing quantitative structure-property relationships (QSPR) that were trained upon the same druglike molecules but two different sets of solubility data: (i) data extracted from several different sources from the published literature, for which the experimental uncertainty is estimated to be 0.6-0.7 log S units (referred to mol/L); (ii) data measured by a single accurate experimental method (CheqSol), for which experimental uncertainty is typically <0.05 log S units. Contrary to what might be expected, the models derived from the CheqSol experimental data are not more accurate than those derived from the "noisy" literature data. The results suggest that, at the present time, it is the deficiency of QSPR methods (algorithms and/or descriptor sets), and not, as is commonly quoted, the uncertainty in the experimental measurements, which is the limiting factor in accurately predicting aqueous solubility for pharmaceutical molecules.

Bioprocess monitoring applications of an innovative ATR-FTIR spectroscopy platform.

Pharmaceutical manufacturing is reliant upon bioprocessing approaches to generate the range of therapeutic products that are available today. The high cost of production, susceptibility to process failure, and requirement to achieve consistent, high-quality product means that process monitoring is paramount during manufacturing. Process analytic technologies (PAT) are key to ensuring high quality product is produced at all stages of development. Spectroscopy-based technologies are well suited as PAT approaches as they are non-destructive and require minimum sample preparation. This study explored the use of a novel attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy platform, which utilises disposable internal reflection elements (IREs), as a method of upstream bioprocess monitoring. The platform was used to characterise organism health and to quantify cellular metabolites in growth media using quantification models to predict glucose and lactic acid levels both singularly and combined. Separation of the healthy and nutrient deficient cells within PC space was clearly apparent, indicating this technique could be used to characterise these classes. For the metabolite quantification, the binary models yielded R 2 values of 0.969 for glucose, 0.976 for lactic acid. When quantifying the metabolites in tandem using a multi-output partial least squares model, the corresponding R 2 value was 0.980. This initial study highlights the suitability of the platform for bioprocess monitoring and paves the way for future in-line developments.