Mario Picciani
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Mario Picciani has a total of 75 co-authors. Here are the top co-authors by number of shared publications:
- Mathias Wilhelm
Computational Mass Spectrometry, School of Life Sciences, Technical University of Munich, Freising, Germany
11 shared publications - Markus List
Data Science in Systems Biology, TUM School of Life Sciences, Technical University of Munich , Freising 85354
6 shared publications - Judith Bernett
Technical University of Munich
4 shared publications - Wassim Gabriel
Computational Mass Spectrometry, School of Life Sciences, Technical University of Munich, Freising, Germany
3 shared publications - Charlotte Adams
University of Antwerp
2 shared publications - Katharina Baum
Freie Universität Berlin, Department of Mathematics and Computer Science, Berlin, Germany
2 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 12 May 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Critical evaluation of drug response prediction models with DrEval
Large-scale drug sensitivity screens have enabled training drug response prediction models based on cancer cell line omics profiles to advance personalized medicine. While model performances reported in the literature appear promising, successful translation to the clinic remains limited. In this work, we discuss key obstacles that lead to overly optimistic performance estimates of state-of-the-art models, making it challenging to track progress in the field. To address them, we present DrEval, a pipeline for unbiased, biologically meaningful evaluation of cancer drug response models. DrEval is designed as a living open-source benchmark that integrates baseline and literature models with standardized hyperparameter tuning, statistically rigorous evaluation, cross-study benchmarks, and supports ablation studies and publication-ready visualizations. Using DrEval, we show that deep learning models barely outperform a naive model that predicts only the mean drug and cell line effects, while no complex model outperforms properly tuned tree-based ensemble baselines in relevant settings.
- 12 May 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Top Cited PapersOpen AccessCritical evaluation of drug response prediction models with DrEval
Large-scale drug sensitivity screens have enabled training drug response prediction models based on cancer cell line omics profiles to advance personalized medicine. While model performances reported in the literature appear promising, successful translation to the clinic remains limited. In this work, we discuss key obstacles that lead to overly optimistic performance estimates of state-of-the-art models, making it challenging to track progress in the field. To address them, we present DrEval, a pipeline for unbiased, biologically meaningful evaluation of cancer drug response models. DrEval is designed as a living open-source benchmark that integrates baseline and literature models with standardized hyperparameter tuning, statistically rigorous evaluation, cross-study benchmarks, and supports ablation studies and publication-ready visualizations. Using DrEval, we show that deep learning models barely outperform a naive model that predicts only the mean drug and cell line effects, while no complex model outperforms properly tuned tree-based ensemble baselines in relevant settings.
- Open Access
- 1 July 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-15
Prosit-XL: enhanced cross-linked peptide identification by fragment intensity prediction to study protein interactions and structures
It has been shown that integrating peptide property predictions such as fragment intensity into the scoring process of peptide spectrum match can greatly increase the number of confidently identified peptides compared to using traditional scoring methods. Here, we introduce Prosit-XL, a robust and accurate fragment intensity predictor covering the cleavable (DSSO/DSBU) and non-cleavable cross-linkers (DSS/BS3), achieving high accuracy on various holdout sets with consistent performance on external datasets without fine-tuning. Due to the complex nature of false positives in XL-MS, an approach to data-driven rescoring was developed that benefits from Prosit-XL’s predictions while limiting the overestimation of the false discovery rate (FDR). After validating this approach using two ground truth datasets consisting of synthetic peptides and proteins, we applied Prosit-XL on a proteome-scale dataset, demonstrating an up to ~3.4-fold improvement in PPI discovery compared to classic approaches. Finally, Prosit-XL was used to increase the coverage and depth of a spatially resolved interactome map of intact human cytomegalovirus virions, leading to the discovery of previously unobserved interactions between human and cytomegalovirus proteins.
- 1 July 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-15
Open AccessProsit-XL: enhanced cross-linked peptide identification by fragment intensity prediction to study protein interactions and structures
It has been shown that integrating peptide property predictions such as fragment intensity into the scoring process of peptide spectrum match can greatly increase the number of confidently identified peptides compared to using traditional scoring methods. Here, we introduce Prosit-XL, a robust and accurate fragment intensity predictor covering the cleavable (DSSO/DSBU) and non-cleavable cross-linkers (DSS/BS3), achieving high accuracy on various holdout sets with consistent performance on external datasets without fine-tuning. Due to the complex nature of false positives in XL-MS, an approach to data-driven rescoring was developed that benefits from Prosit-XL’s predictions while limiting the overestimation of the false discovery rate (FDR). After validating this approach using two ground truth datasets consisting of synthetic peptides and proteins, we applied Prosit-XL on a proteome-scale dataset, demonstrating an up to ~3.4-fold improvement in PPI discovery compared to classic approaches. Finally, Prosit-XL was used to increase the coverage and depth of a spatially resolved interactome map of intact human cytomegalovirus virions, leading to the discovery of previously unobserved interactions between human and cytomegalovirus proteins.
- Open Access
From Hype to Health Check: Critical Evaluation of Drug Response Prediction Models with DrEval
Motivation: Large-scale drug sensitivity screens have enabled training drug response prediction models based on cancer cell line omics profiles to facilitate personalized medicine. While model performances reported in the literature appear promising, no successful translation to the clinic has been reported. Results: We identify six primary obstacles to this: non-reproducible models, data leakage leading to poor generalization, pseudoreplication, biased evaluation metrics, missing ablation studies, and inconsistent viability data reducing comparability across models. Together, these issues lead to overly optimistic performance estimates of state-of-the-art models and make it challenging to track progress in the field. To address this, we present DrEval, a pipeline for unbiased and biologically meaningful evaluation of cancer drug response models. It includes baseline and literature models with consistent hyperparameter tuning, statistically sound evaluations, and cross-study benchmarks. DrEval enables ablation studies and publication-ready visualizations. It allows researchers to focus on model development without implementing their own evaluation protocol. We find that deep learning models barely outperform a naive model predicting the mean drug and cell line effects, while no complex model significantly outperforms properly tuned tree-based ensemble baselines in relevant settings. We advocate making our pipeline a standard benchmark for cancer drug response prediction, ensuring a clinically relevant and robust assessment. Availability and implementation: DrEval consists of a Python package, available on PyPI (drevalpy) and GitHub ( github.com/daisybio/drevalpy ), and an accompanying nf-core pipeline ( github.com/nf-core/drugresponseeval ). All data is available on Zenodo (DOI: 10.5281/zenodo.12633909), preprocessing scripts on github.com/daisybio/preprocess_drp_data .
Open AccessFrom Hype to Health Check: Critical Evaluation of Drug Response Prediction Models with DrEval
Motivation: Large-scale drug sensitivity screens have enabled training drug response prediction models based on cancer cell line omics profiles to facilitate personalized medicine. While model performances reported in the literature appear promising, no successful translation to the clinic has been reported. Results: We identify six primary obstacles to this: non-reproducible models, data leakage leading to poor generalization, pseudoreplication, biased evaluation metrics, missing ablation studies, and inconsistent viability data reducing comparability across models. Together, these issues lead to overly optimistic performance estimates of state-of-the-art models and make it challenging to track progress in the field. To address this, we present DrEval, a pipeline for unbiased and biologically meaningful evaluation of cancer drug response models. It includes baseline and literature models with consistent hyperparameter tuning, statistically sound evaluations, and cross-study benchmarks. DrEval enables ablation studies and publication-ready visualizations. It allows researchers to focus on model development without implementing their own evaluation protocol. We find that deep learning models barely outperform a naive model predicting the mean drug and cell line effects, while no complex model significantly outperforms properly tuned tree-based ensemble baselines in relevant settings. We advocate making our pipeline a standard benchmark for cancer drug response prediction, ensuring a clinically relevant and robust assessment. Availability and implementation: DrEval consists of a Python package, available on PyPI (drevalpy) and GitHub ( github.com/daisybio/drevalpy ), and an accompanying nf-core pipeline ( github.com/nf-core/drugresponseeval ). All data is available on Zenodo (DOI: 10.5281/zenodo.12633909), preprocessing scripts on github.com/daisybio/preprocess_drp_data .
- Open Access
To fly, or not to fly, that is the question: A deep learning model for peptide detectability prediction in mass spectrometry
Identifying detectable peptides, known as flyers, is key in mass spectrometry-based proteomics. Peptide detectability is strongly related with the peptide sequence and its resulting physicochemical properties. Moreover, the high variability in MS data, particularly in peptide detectability and intensity across multiple analyses and samples, makes the development of a generic model for detectability prediction unfeasible. This underlines the need for tools that can be refined for specific experimental conditions. To address this need, we present Pfly, a deep learning model developed to predicts peptide detectability based solely on peptide sequence. Pfly distinguishes itself as a versatile and reliable state-of-the-art tool, offering high performance, accessibility, and easy customizability for end-users. This adaptability allows researchers to tailor the model to their specific experimental conditions, facilitating the creation of lab-specific models. This, in turn, can lead to more accurate results and expand the model’s applicability across various research fields. The model’s architecture is an encoder-decoder with an attention mechanism. This tool classifies peptides as either flyers or non-flyers, providing both binary probabilities and detailed categorical probabilities for four distinct classes defined in this study: non-flyer, weak flyer, intermediate flyer, and strong flyer. The model was initially trained on a synthetic peptide library and subsequently fine-tuned with a biological dataset to mitigate bias towards synthesizability, improving the predictive capacity and outperforming state-of-the-art predictors in a benchmark comparison. The study further investigates the influence of protein abundance and the search engine, illustrating the negative impact on peptide identification due to misclassification. Pfly has been integrated in the DLOmix framework and it is accessible on GitHub at https://github.com/wilhelm-lab/dlomix .
Open AccessTo fly, or not to fly, that is the question: A deep learning model for peptide detectability prediction in mass spectrometry
Identifying detectable peptides, known as flyers, is key in mass spectrometry-based proteomics. Peptide detectability is strongly related with the peptide sequence and its resulting physicochemical properties. Moreover, the high variability in MS data, particularly in peptide detectability and intensity across multiple analyses and samples, makes the development of a generic model for detectability prediction unfeasible. This underlines the need for tools that can be refined for specific experimental conditions. To address this need, we present Pfly, a deep learning model developed to predicts peptide detectability based solely on peptide sequence. Pfly distinguishes itself as a versatile and reliable state-of-the-art tool, offering high performance, accessibility, and easy customizability for end-users. This adaptability allows researchers to tailor the model to their specific experimental conditions, facilitating the creation of lab-specific models. This, in turn, can lead to more accurate results and expand the model’s applicability across various research fields. The model’s architecture is an encoder-decoder with an attention mechanism. This tool classifies peptides as either flyers or non-flyers, providing both binary probabilities and detailed categorical probabilities for four distinct classes defined in this study: non-flyer, weak flyer, intermediate flyer, and strong flyer. The model was initially trained on a synthetic peptide library and subsequently fine-tuned with a biological dataset to mitigate bias towards synthesizability, improving the predictive capacity and outperforming state-of-the-art predictors in a benchmark comparison. The study further investigates the influence of protein abundance and the search engine, illustrating the negative impact on peptide identification due to misclassification. Pfly has been integrated in the DLOmix framework and it is accessible on GitHub at https://github.com/wilhelm-lab/dlomix .
- Open Access
- 23 August 2024
- Published by Oxford University Press (OUP) in Nucleic Acids Research
- Vol. 52 (17) , 10144-10160
Network medicine-based epistasis detection in complex diseases: ready for quantum computing
Most heritable diseases are polygenic. To comprehend the underlying genetic architecture, it is crucial to discover the clinically relevant epistatic interactions (EIs) between genomic single nucleotide polymorphisms (SNPs) ( 1–3). Existing statistical computational methods for EI detection are mostly limited to pairs of SNPs due to the combinatorial explosion of higher-order EIs. With NeEDL (network-based epistasis detection via local search), we leverage network medicine to inform the selection of EIs that are an order of magnitude more statistically significant compared to existing tools and consist, on average, of five SNPs. We further show that this computationally demanding task can be substantially accelerated once quantum computing hardware becomes available. We apply NeEDL to eight different diseases and discover genes (affected by EIs of SNPs) that are partly known to affect the disease, additionally, these results are reproducible across independent cohorts. EIs for these eight diseases can be interactively explored in the Epistasis Disease Atlas (https://epistasis-disease-atlas.com). In summary, NeEDL demonstrates the potential of seamlessly integrated quantum computing techniques to accelerate biomedical research. Our network medicine approach detects higher-order EIs with unprecedented statistical and biological evidence, yielding unique insights into polygenic diseases and providing a basis for the development of improved risk scores and combination therapies.
- 23 August 2024
- Published by Oxford University Press (OUP) in Nucleic Acids Research
- Vol. 52 (17) , 10144-10160
Open AccessNetwork medicine-based epistasis detection in complex diseases: ready for quantum computing
Most heritable diseases are polygenic. To comprehend the underlying genetic architecture, it is crucial to discover the clinically relevant epistatic interactions (EIs) between genomic single nucleotide polymorphisms (SNPs) ( 1–3). Existing statistical computational methods for EI detection are mostly limited to pairs of SNPs due to the combinatorial explosion of higher-order EIs. With NeEDL (network-based epistasis detection via local search), we leverage network medicine to inform the selection of EIs that are an order of magnitude more statistically significant compared to existing tools and consist, on average, of five SNPs. We further show that this computationally demanding task can be substantially accelerated once quantum computing hardware becomes available. We apply NeEDL to eight different diseases and discover genes (affected by EIs of SNPs) that are partly known to affect the disease, additionally, these results are reproducible across independent cohorts. EIs for these eight diseases can be interactively explored in the Epistasis Disease Atlas (https://epistasis-disease-atlas.com). In summary, NeEDL demonstrates the potential of seamlessly integrated quantum computing techniques to accelerate biomedical research. Our network medicine approach detects higher-order EIs with unprecedented statistical and biological evidence, yielding unique insights into polygenic diseases and providing a basis for the development of improved risk scores and combination therapies.
- Open Access
- 10 May 2024
- Published by Springer Nature in Nature Communications
- Vol. 15 (1) , 1-11
Fragment ion intensity prediction improves the identification rate of non-tryptic peptides in timsTOF
Immunopeptidomics is crucial for immunotherapy and vaccine development. Because the generation of immunopeptides from their parent proteins does not adhere to clear-cut rules, rather than being able to use known digestion patterns, every possible protein subsequence within human leukocyte antigen (HLA) class-specific length restrictions needs to be considered during sequence database searching. This leads to an inflation of the search space and results in lower spectrum annotation rates. Peptide-spectrum match (PSM) rescoring is a powerful enhancement of standard searching that boosts the spectrum annotation performance. We analyze 302,105 unique synthesized non-tryptic peptides from the ProteomeTools project on a timsTOF-Pro to generate a ground-truth dataset containing 93,227 MS/MS spectra of 74,847 unique peptides, that is used to fine-tune the deep learning-based fragment ion intensity prediction model Prosit. We demonstrate up to 3-fold improvement in the identification of immunopeptides, as well as increased detection of immunopeptides from low input samples.
- 10 May 2024
- Published by Springer Nature in Nature Communications
- Vol. 15 (1) , 1-11
Open AccessFragment ion intensity prediction improves the identification rate of non-tryptic peptides in timsTOF
Immunopeptidomics is crucial for immunotherapy and vaccine development. Because the generation of immunopeptides from their parent proteins does not adhere to clear-cut rules, rather than being able to use known digestion patterns, every possible protein subsequence within human leukocyte antigen (HLA) class-specific length restrictions needs to be considered during sequence database searching. This leads to an inflation of the search space and results in lower spectrum annotation rates. Peptide-spectrum match (PSM) rescoring is a powerful enhancement of standard searching that boosts the spectrum annotation performance. We analyze 302,105 unique synthesized non-tryptic peptides from the ProteomeTools project on a timsTOF-Pro to generate a ground-truth dataset containing 93,227 MS/MS spectra of 74,847 unique peptides, that is used to fine-tune the deep learning-based fragment ion intensity prediction model Prosit. We demonstrate up to 3-fold improvement in the identification of immunopeptides, as well as increased detection of immunopeptides from low input samples.
- Open Access
- 8 May 2024
- Published by ScienceOpen
The good, the bad, and the ugly: Drug response prediction using machine learning
Drug response prediction using molecular readouts (e.g., transcriptomics) is an active area of research due to its potentially pivotal role in personalized medicine. In theory, it would allow treating physicians to tailor recommendations to individuals. Machine learning has emerged as a promising method for predicting complex outcomes and thus has been used in drug response prediction. We have shown that state-of-the-art models suffer from various problems in the past. Here, we provide an update on this and aim to address whether drug response prediction using machine learning is currently possible and what issues remain to be solved to allow model development and generalization. For this, we cover various aspects of the machine learning life cycle, from available training data to generalization. Briefly, we show that currently available data shows substantial levels of heterogeneity, often enriched in specific cell lines and drugs. This questions the use of independent hold-out data for model performance assessment. To circumvent high noise levels and outliers in the training data – a result of high throughput screening efforts – we developed a foundational model using self-supervised learning that can be used to model drug response prediction with and embedding that is aware of the full concentration range rather than relying on simplified summary statistics. Last, we present our efforts to develop a ready-to-use pipeline for model training, testing, and evaluation that comes with several large curated and normalized training data and discuss whether the available data and complexity of the problem allows training drug response prediction models showcased on preliminary data where several dimensionality reduction methods were used to reduce the p>n challenge.
- 8 May 2024
- Published by ScienceOpen
Open AccessThe good, the bad, and the ugly: Drug response prediction using machine learning
Drug response prediction using molecular readouts (e.g., transcriptomics) is an active area of research due to its potentially pivotal role in personalized medicine. In theory, it would allow treating physicians to tailor recommendations to individuals. Machine learning has emerged as a promising method for predicting complex outcomes and thus has been used in drug response prediction. We have shown that state-of-the-art models suffer from various problems in the past. Here, we provide an update on this and aim to address whether drug response prediction using machine learning is currently possible and what issues remain to be solved to allow model development and generalization. For this, we cover various aspects of the machine learning life cycle, from available training data to generalization. Briefly, we show that currently available data shows substantial levels of heterogeneity, often enriched in specific cell lines and drugs. This questions the use of independent hold-out data for model performance assessment. To circumvent high noise levels and outliers in the training data – a result of high throughput screening efforts – we developed a foundational model using self-supervised learning that can be used to model drug response prediction with and embedding that is aware of the full concentration range rather than relying on simplified summary statistics. Last, we present our efforts to develop a ready-to-use pipeline for model training, testing, and evaluation that comes with several large curated and normalized training data and discuss whether the available data and complexity of the problem allows training drug response prediction models showcased on preliminary data where several dimensionality reduction methods were used to reduce the p>n challenge.
- Open Access
- 6 September 2023
- Published by Wiley in Proteomics
- Vol. 24 (8) , e2300112
Oktoberfest: Open‐source spectral library generation and rescoring pipeline based on Prosit
Machine learning (ML) and deep learning (DL) models for peptide property prediction such as Prosit have enabled the creation of high quality in silico reference libraries. These libraries are used in various applications, ranging from data‐independent acquisition (DIA) data analysis to data‐driven rescoring of search engine results. Here, we present Oktoberfest, an open source Python package of our spectral library generation and rescoring pipeline originally only available online via ProteomicsDB. Oktoberfest is largely search engine agnostic and provides access to online peptide property predictions, promoting the adoption of state‐of‐the‐art ML/DL models in proteomics analysis pipelines. We demonstrate its ability to reproduce and even improve our results from previously published rescoring analyses on two distinct use cases. Oktoberfest is freely available on GitHub (https://github.com/wilhelm‐lab/oktoberfest) and can easily be installed locally through the cross‐platform PyPI Python package.
- 6 September 2023
- Published by Wiley in Proteomics
- Vol. 24 (8) , e2300112
Open AccessOktoberfest: Open‐source spectral library generation and rescoring pipeline based on Prosit
Machine learning (ML) and deep learning (DL) models for peptide property prediction such as Prosit have enabled the creation of high quality in silico reference libraries. These libraries are used in various applications, ranging from data‐independent acquisition (DIA) data analysis to data‐driven rescoring of search engine results. Here, we present Oktoberfest, an open source Python package of our spectral library generation and rescoring pipeline originally only available online via ProteomicsDB. Oktoberfest is largely search engine agnostic and provides access to online peptide property predictions, promoting the adoption of state‐of‐the‐art ML/DL models in proteomics analysis pipelines. We demonstrate its ability to reproduce and even improve our results from previously published rescoring analyses on two distinct use cases. Oktoberfest is freely available on GitHub (https://github.com/wilhelm‐lab/oktoberfest) and can easily be installed locally through the cross‐platform PyPI Python package.
- Open Access
Fragment ion intensity prediction improves the identification rate of non-tryptic peptides in timsTOF
Immunopeptidomics plays a crucial role in identifying targets for immunotherapy and vaccine development. Because the generation of immunopeptides from their parent proteins does not adhere to clear-cut rules, rather than being able to use known digestion patterns, every possible protein subsequence within human leukocyte antigen (HLA) class-specific length restrictions needs to be considered during sequence database searching. This leads to an inflation of the search space and results in lower spectrum annotation rates. Peptide-spectrum match (PSM) rescoring is a powerful enhancement of standard searching that boosts the spectrum annotation performance. Low abundant peptides often occur in the field of immunopeptidomics, which is why the highly sensitive timsTOF instruments are increasingly gaining popularity. To improve PSM rescoring for immunopeptides measured using timsTOF instruments, we trained a deep learning-based fragment ion intensity prediction model. 302,105 unique synthesized non-tryptic peptides from the ProteomeTools project were analyzed on a timsTOF-Pro to generate a ground-truth dataset, containing 93,227 MS/MS spectra of 74,847 unique peptides, that was used to fine-tune an existing Prosit model. By applying our fragment ion intensity prediction model, we demonstrate up to 3-fold improvement in the identification of immunopeptides. Furthermore, our approach increased detection of immunopeptides even from low input samples.
Open AccessFragment ion intensity prediction improves the identification rate of non-tryptic peptides in timsTOF
Immunopeptidomics plays a crucial role in identifying targets for immunotherapy and vaccine development. Because the generation of immunopeptides from their parent proteins does not adhere to clear-cut rules, rather than being able to use known digestion patterns, every possible protein subsequence within human leukocyte antigen (HLA) class-specific length restrictions needs to be considered during sequence database searching. This leads to an inflation of the search space and results in lower spectrum annotation rates. Peptide-spectrum match (PSM) rescoring is a powerful enhancement of standard searching that boosts the spectrum annotation performance. Low abundant peptides often occur in the field of immunopeptidomics, which is why the highly sensitive timsTOF instruments are increasingly gaining popularity. To improve PSM rescoring for immunopeptides measured using timsTOF instruments, we trained a deep learning-based fragment ion intensity prediction model. 302,105 unique synthesized non-tryptic peptides from the ProteomeTools project were analyzed on a timsTOF-Pro to generate a ground-truth dataset, containing 93,227 MS/MS spectra of 74,847 unique peptides, that was used to fine-tune an existing Prosit model. By applying our fragment ion intensity prediction model, we demonstrate up to 3-fold improvement in the identification of immunopeptides. Furthermore, our approach increased detection of immunopeptides even from low input samples.
- Open Access
- 1 January 2023
- Published by ScienceOpen
Pitfalls of computational drug response prediction
Drug response prediction is pivotal in personalized medicine as it allows healthcare providers to tailor treatments to individual patients for improved therapeutic outcomes. Machine learning has emerged as a promising tool for predicting drug responses; however, this burgeoning field faces three substantial challenges that remain inadequately addressed. State-of-the-art methods, including deep learning approaches, do not systematically compare themselves against simple baseline models. We found that published methods lack reproducible implementations, suffer from overfitting, and do not pass fundamental consistency and robustness tests. This leads to over-optimism about prediction results, which can often be matched by a simple mean predictor. In addition, frequently used measures to quantify and summarize drug response, such as IC50, EC50, and AUC values, do not capture the dynamic of drug response adequately, leading to poor generalization of findings across data sets. Finally, drug response prediction methods typically rely exclusively on transcriptomics measurements, neglecting proteomics and phospho-proteomics data, which offer unique insights into the response of the drug targets themselves. To address these research gaps, we propose the development of a new pipeline for systematically benchmarking drug response prediction methods against baseline and state-of-the-art predictors on curated multi-omics data in ProteomicsDB. Through these concerted efforts, we pave the way for more accurate and clinically relevant drug response predictions, ultimately enhancing patient care and treatment outcomes.
- 1 January 2023
- Published by ScienceOpen
Open AccessPitfalls of computational drug response prediction
Drug response prediction is pivotal in personalized medicine as it allows healthcare providers to tailor treatments to individual patients for improved therapeutic outcomes. Machine learning has emerged as a promising tool for predicting drug responses; however, this burgeoning field faces three substantial challenges that remain inadequately addressed. State-of-the-art methods, including deep learning approaches, do not systematically compare themselves against simple baseline models. We found that published methods lack reproducible implementations, suffer from overfitting, and do not pass fundamental consistency and robustness tests. This leads to over-optimism about prediction results, which can often be matched by a simple mean predictor. In addition, frequently used measures to quantify and summarize drug response, such as IC50, EC50, and AUC values, do not capture the dynamic of drug response adequately, leading to poor generalization of findings across data sets. Finally, drug response prediction methods typically rely exclusively on transcriptomics measurements, neglecting proteomics and phospho-proteomics data, which offer unique insights into the response of the drug targets themselves. To address these research gaps, we propose the development of a new pipeline for systematically benchmarking drug response prediction methods against baseline and state-of-the-art predictors on curated multi-omics data in ProteomicsDB. Through these concerted efforts, we pave the way for more accurate and clinically relevant drug response predictions, ultimately enhancing patient care and treatment outcomes.