Machine learning to predict continuous protein properties from binary cell sorting data and map unseen sequence space

Proteins are a diverse class of biomolecules responsible for wide-ranging cellular functions, from catalyzing reactions to recognizing pathogens. The ability to evolve proteins rapidly and inexpensively toward improved properties is a common objective for protein engineers. Powerful high-throughput methods like fluorescent activated cell sorting and next-generation sequencing have dramatically improved directed evolution experiments. However, it is unclear how to best leverage these data to characterize protein fitness landscapes more completely and identify lead candidates. In this work, we develop a simple yet powerful framework to improve protein optimization by predicting continuous protein properties from simple directed evolution experiments using interpretable, linear machine learning models. Importantly, we find that these models, which use data from simple but imprecise experimental estimates of protein fitness, have predictive capabilities that approach more precise but expensive data. Evaluated across five diverse protein engineering tasks, continuous properties are consistently predicted from readily available deep sequencing data, demonstrating that protein fitness space can be reasonably well modeled by linear relationships among sequence mutations. To prospectively test the utility of this approach, we generated a library of stapled peptides and applied the framework to predict affinity and specificity from simple cell sorting data. We then coupled integer linear programming, a method to optimize protein fitness from linear weights, with mutation scores from machine learning to identify variants in unseen sequence space that have improved and co-optimal properties. This approach represents a versatile tool for improved analysis and identification of protein variants across many domains of protein engineering.

Rapid Evaluation of Staple Placement in Stabilized α Helices Using Bacterial Surface Display

There are a wealth of proteins involved in disease that cannot be targeted by current therapeutics because they are inside cells, inaccessible to most macromolecules, and lack small-molecule binding pockets. Stapled peptides, where two amino acids are covalently linked, form a class of macrocycles that have the potential to penetrate cell membranes and disrupt intracellular protein-protein interactions. However, their discovery relies on solid-phase synthesis, greatly limiting queries into their complex design space involving amino acid sequence, staple location, and staple chemistry. Here, we use stabilized peptide engineering by Escherichia coli display (SPEED), which utilizes noncanonical amino acids and click chemistry for stabilization, to rapidly screen staple location and linker structure to accelerate peptide design. After using SPEED to confirm hotspots in the mdm2-p53 interaction, we evaluated different staple locations and staple chemistry to identify several novel nanomolar and sub-nanomolar antagonists. Next, we evaluated SPEED in the B cell lymphoma 2 (Bcl-2) protein family, which is responsible for regulating apoptosis. We report that novel staple locations modified in the context of BIM, a high affinity but nonspecific naturally occurring peptide, improve its specificity against the highly homologous proteins in the Bcl-2 family. These compounds demonstrate the importance of screening linker location and chemistry in identifying high affinity and specific peptide antagonists. Therefore, SPEED can be used as a versatile platform to evaluate multiple design criteria for stabilized peptide engineering.

Directed Evolution Using Stabilized Bacterial Peptide Display

Chemically stabilized peptides have attracted intense interest by academics and pharmaceutical companies due to their potential to hit currently "undruggable" targets. However, engineering an optimal sequence, stabilizing linker location, and physicochemical properties is a slow and arduous process. By pairing non-natural amino acid incorporation and cell surface click chemistry in bacteria with high-throughput sorting, we developed a method to quantitatively select high affinity ligands and applied the Stabilized Peptide Evolution by E. coli Display technique to develop disrupters of the therapeutically relevant MDM2-p53 interface. Through in situ stabilization on the bacterial surface, we demonstrate rapid isolation of stabilized peptides with improved affinity and novel structures. Several peptides evolved a second loop including one sequence (K_d = 1.8 nM) containing an i, i+4 disulfide bond. NMR structural determination indicated a bent helix in solution and bound to MDM2. The bicyclic peptide had improved protease stability, and we demonstrated that protease resistance could be measured both on the bacterial surface and in solution, enabling the method to test and/or screen for additional drug-like properties critical for biologically active compounds.

A phase I/II trial of AT9283, a selective inhibitor of aurora kinase in children with relapsed or refractory acute leukemia: challenges to run early phase clinical trials for children with leukemia

Aurora kinases regulate mitosis and are commonly overexpressed in leukemia. This phase I/IIa study of AT9283, a multikinase inhibitor, was designed to identify maximal tolerated doses, safety, pharmacokinetics, and pharmacodynamic activity in children with relapsed/refractory acute leukemia. The trial suffered from poor recruitment and terminated early, therefore failing to identify its primary endpoints. AT9283 caused tolerable toxicity, but failed to show clinical responses. Future trials should be based on robust preclinical data that provide an indication of which patients may benefit from the experimental agent, and recruitment should be improved through international collaborations and early combination with established treatment strategies.

Longitudinal Trajectories of Clinical Decline in Amyloid Positive and Negative Populations

BACKGROUND

Brain beta-amyloid status portends different trajectories of clinical decline.

OBJECTIVE

Determine trajectories and predictive baseline variable(s).

DESIGN

Longitudinal, up to 24 months.

SETTING

ADNI sites.

PARTICIPANTS

Healthy control (n=325), early and late mild cognitive impairment (n=279; n=372), and Alzheimer's dementia (n=216) subjects from ADNI-1/GO/2.

MEASUREMENTS

Baseline amyloid status was based on first available CSF Aβ1-42 or, [11C]PiB or [18F]florbetapir (FBP) PET. Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-Cog13) and Functional Activities Questionnaire (FAQ) were co-analyzed using Growth Mixture Modeling (GMM) to define latent class trajectories for each amyloid group. Classification and Regression Tree (CART) analysis determined which variables best predicted trajectory class membership using a number of variables available to clinicians.

RESULTS

GMMs found two trajectory classes (C1, C2) each for amyloid-positive (P; n=722) and negative (N; n=470) groups. Most (90%) in the negative group were C2N with mildly impaired baseline ADAS-Cog13, normal FAQ and nonprogression; 10% were C1N with moderately impaired baseline FAQ and ADAS-Cog13 and trajectory of moderately worsening scores on the FAQ. C1P (26%) had more impaired baseline FAQ and ADAS-Cog13 than C2P (74%) and a steeper declining trajectory. CART yielded 4 decision nodes (FAQ <10.5, FAQ <6.5, MMSE ≥26.5, age <75.5) in positive and 1 node (FAQ <6.5) in negative groups, with 91.4% and 92.8% accuracy for class assignments, respectively.

CONCLUSIONS

The trajectory pattern of greater decline in amyloid positive subjects was predicted by greater baseline impairment of cognition and function. While most amyloid-negative subjects had nonprogression irrespective of their diagnosis, a subgroup declined similarly to the gradually declining amyloid-positive group. CART predicted likely trajectory class, with known amyloid status, using variables accessible in a clinical setting, but needs replication.

The heterogeneity of antipsychotic response in the treatment of schizophrenia

BACKGROUND

Schizophrenia is a heterogeneous disorder in terms of patient response to antipsychotic treatment. Understanding the heterogeneity of treatment response may help to guide treatment decisions. This study was undertaken to capture inherent patterns of response to antipsychotic treatment in patients with schizophrenia, characterize the subgroups of patients with similar courses of response, and examine illness characteristics at baseline as possible predictors of response.

METHOD

Growth mixture modeling (GMM) was applied to data from a randomized, double-blind, 12-week study of 628 patients with schizophrenia or schizo-affective disorder treated with risperidone or olanzapine.

RESULTS

Four distinct response trajectories based on Positive and Negative Syndrome Scale (PANSS) total score over 12 weeks were identified: Class 1 (420 patients, 80.6%) with moderate average baseline PANSS total score showing gradual symptom improvement; Class 2 (65 patients, 12.5%) showing rapid symptom improvement; Class 3 (24 patients, 4.6%) with high average baseline PANSS total score showing gradual symptom improvement; and Class 4 (12 patients, 2.3%) showing unsustained symptom improvement. Latent class membership of early responders (ER) and early non-responders (ENR) was determined based on 20% symptom improvement criteria at 2 weeks and ultimate responders (UR) and ultimate non-responders (UNR) based on 40% symptom improvement criteria at 12 weeks. Baseline factors with potential influence on latent class membership were identified.

CONCLUSIONS

This study identified four distinct treatment response patterns with predominant representation of responders or non-responders to treatment in these classes. This heterogeneity may represent discrete endophenotypes of response to treatment with different etiologic underpinnings.

Systems biology of the neurospora biological clock

A major challenge of systems biology is explaining complex traits, such as the biological clock, in terms of the kinetics of macromolecules. The clock poses at least four challenges for systems biology: (i) identifying the genetic network to explain the clock mechanism quantitatively; (ii) specifying the clock's functional connection to a thousand or more genes and their products in the genome; (iii) explaining the clock's response to light and other environmental cues; and (iv) explaining how the clock's genetic network evolves. Here, the authors illustrate an approach to these problems by fitting an ensemble of genetic networks to microarray data derived from oligonucleotide arrays with approximately all 11 000 Neurospora crassa genes represented. A promising genetic network for the clock mechanism is identified.

Pancreatic ductal bicarbonate secretion: past, present and future

The pancreatic duct epithelium in the guinea-pig and many other species secretes HCO(3)(-) at concentrations approaching 150 mM. This cannot be explained by conventional models based upon HCO(3)(-) secretion via an anion exchanger at the luminal membrane because: 1) under these conditions, the Cl(-) and HCO(3)(-) concentration gradients would favour HCO(3)(-) reabsorption rather than secretion, and 2) the luminal anion exchanger appears to be inhibited by luminal HCO(3)(-) concentrations of 125 mM or more. There may, however, be a sufficiently large electrochemical gradient to drive HCO(3)(-) secretion across the luminal membrane via an anion conductance. In contrast to earlier studies on rat ducts, the membrane potential E(m) in guinea-pig duct cells does not depolarise appreciably upon stimulation with secretagogues but remains constant at about -60 mV. Consequently, even with 125 mM or more HCO(3)(-) in the lumen and an estimated 20 mM in the cytoplasm, the electrochemical gradient for HCO(3)(-) will still favour secretion to the lumen. Under the same conditions, the intracellular Cl(-) concentration drops to very low levels (approximately 7 mM) presumably because, although Cl(-) may leave freely through the cystic fibrosis transmembrane conductance regulator (CFTR) channels in the luminal membrane, there is no major pathway for Cl(-) uptake across the basolateral membrane. Consequently a HCO(3)(-)-rich secretion may arise as a result of the lack of competition from intracellular Cl(-) for efflux via the anion conductances at the luminal membrane. Whether CFTR, or another anion conductance, provides such a pathway for HCO(3)(-) remains to be seen.

Unsafe sleep practices and an analysis of bedsharing among infants dying suddenly and unexpectedly: results of a four-year, population-based, death-scene investigation study of sudden infant death syndrome and related deaths

BACKGROUND

Prone sleep and unsafe sleep surfaces increase the risk of sudden infant death. Recent epidemiologic studies also suggest that when an infant's head or face is covered by bedding, or when a sleep surface is shared with others, the risk of dying increases. The inference of a causal role for these risk factors is supported by physiologic studies and by the consistent finding that fewer infants die when risk factors are reduced. The prevalence of most of these risk factors in infant deaths in the United States is uncertain.

OBJECTIVE

To describe the prevalence of several important risk factors related to sleep practices among a defined population of infants dying suddenly and unexpectedly.

METHODS

In this population-based study, we retrospectively reviewed death-scene information and medical examiners' investigations of deaths in the city of St Louis and St Louis County between January 1, 1994 and December 31, 1997. Because of the potential for diagnostic overlap, all deaths involving infants <2 years old with the diagnoses of sudden infant death syndrome (SIDS), accidental suffocation, or cause undetermined were included.

RESULTS

The deaths of 119 infants were studied. Their mean age was 109.3 days (range: 6-350). The diagnoses were SIDS in 88 deaths, accidental suffocation in 16, and undetermined in 15. Infants were found prone in 61.1% of cases and were found on a sleep surface not designed for infants in 75.9%. The head or face was covered by bedding in 29.4%. A shared sleep surface was the site of death in 47.1%. Only 8.4% of deaths involved infants found nonprone and alone, with head and face uncovered.

CONCLUSIONS

Using detailed death-scene descriptions, we found that similar unsafe sleeping practices occurred in the large majority of cases diagnosed as SIDS, accidental suffocation, and cause undetermined. Considering these diagnoses together may be useful in public health campaigns during a time when there may be diagnostic overlap. Regardless of the diagnosis, recommendations that infants sleep supine on firm sleep surfaces that lessen the risk of entrapment or head covering have the potential to save many lives. Campaigns are needed to heighten awareness of these messages and of the risks of dangerous bedsharing.

Increased cerebrospinal fluid 5-hydroxytryptamine and 5-hydroxyindoleacetic acid in Kleine-Levin syndrome

A 17-year-old man with the Kleine-Levin syndrome died unexpectedly of cardiopulmonary arrest during a period of autonomic instability that followed an episode of megaphagia. At autopsy, the only pertinent finding was mild depigmentation of the locus ceruleus and substantia nigra. Premortem CSF levels of 5-hydroxytryptamine (5-HT) and 5-hydroxyindoleacetic acid (5-HIAA) levels were elevated. These findings indicate that many symptoms of the Kleine-Levin syndrome are a result of a neurotransmitter imbalance in the serotonergic pathway of the brainstem.